{
 "name": "llm-speed measured LLM inference benchmarks",
 "version": "2026-10-17.3",
 "generated_at": "2026-10-17T10:04:08Z",
 "license": "CC BY 4.0",
 "license_url": "https://creativecommons.org/licenses/by/4.0/",
 "source": "https://llm-speed.com/data",
 "citation": "llm-speed. Measured LLM inference benchmarks, version 2026-10-17.3. https://llm-speed.com/data. CC BY 4.0.",
 "fields": {
  "run_id": "Run ID; the run page is /r/<run_id>",
  "run_url": "Run page URL",
  "date": "Date the run was received (UTC, YYYY-MM-DD)",
  "hardware": "Hardware summary as reported by the benchmark CLI (accelerator + CPU + RAM)",
  "model": "Model name as reported by the runtime",
  "model_size": "Parameter count as reported by the runtime, if any",
  "quant": "Quantization as reported by the runtime, if any",
  "runtime": "Inference runtime (backend), e.g. llama.cpp, ollama, mlx",
  "runtime_version": "Runtime version, if reported",
  "workload": "Suite workload, e.g. chat-short, agent-trace",
  "suite_version": "Benchmark suite version",
  "decode_tps": "Measured decode speed, tokens/second",
  "prefill_tps": "Measured prompt-processing speed, tokens/second",
  "ttft_ms": "Measured time to first token, ms",
  "prompt_tokens": "Prompt tokens",
  "output_tokens": "Generated tokens",
  "batch_size": "Concurrent sequences",
  "label": "Always 'measured': every row is a signed measurement"
 },
 "run_count": 277,
 "row_count": 522,
 "excluded_flagged_runs": [
  "r_pqjsvd-cub4"
 ],
 "rows": [
  {
   "run_id": "r_xnr_k4emoyr",
   "run_url": "https://llm-speed.com/r/r_xnr_k4emoyr",
   "date": "2026-10-17",
   "hardware": "RTX 5090 (32GB) + AMD EPYC 7543 32-Core Processor (120c) + 945GB",
   "model": "Qwen2.5-Coder-14B-Instruct",
   "model_size": null,
   "quant": "Q4_K_M",
   "runtime": "llama.cpp",
   "runtime_version": "0.7.0-dev (build 1, commit 1268eda)",
   "workload": "concurrent-decode",
   "suite_version": "suite-v1",
   "decode_tps": 136.67,
   "prefill_tps": null,
   "ttft_ms": null,
   "prompt_tokens": 1024,
   "output_tokens": 256,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_xnr_k4emoyr",
   "run_url": "https://llm-speed.com/r/r_xnr_k4emoyr",
   "date": "2026-10-17",
   "hardware": "RTX 5090 (32GB) + AMD EPYC 7543 32-Core Processor (120c) + 945GB",
   "model": "Qwen2.5-Coder-14B-Instruct",
   "model_size": null,
   "quant": "Q4_K_M",
   "runtime": "llama.cpp",
   "runtime_version": "0.7.0-dev (build 1, commit 1268eda)",
   "workload": "chat-short",
   "suite_version": "suite-v1",
   "decode_tps": 137.25,
   "prefill_tps": 2016.91,
   "ttft_ms": 65.0,
   "prompt_tokens": 131,
   "output_tokens": 256,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_xnr_k4emoyr",
   "run_url": "https://llm-speed.com/r/r_xnr_k4emoyr",
   "date": "2026-10-17",
   "hardware": "RTX 5090 (32GB) + AMD EPYC 7543 32-Core Processor (120c) + 945GB",
   "model": "Qwen2.5-Coder-14B-Instruct",
   "model_size": null,
   "quant": "Q4_K_M",
   "runtime": "llama.cpp",
   "runtime_version": "0.7.0-dev (build 1, commit 1268eda)",
   "workload": "chat-long",
   "suite_version": "suite-v1",
   "decode_tps": 128.9,
   "prefill_tps": 5987.2,
   "ttft_ms": 529.1,
   "prompt_tokens": 3168,
   "output_tokens": 564,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_xnr_k4emoyr",
   "run_url": "https://llm-speed.com/r/r_xnr_k4emoyr",
   "date": "2026-10-17",
   "hardware": "RTX 5090 (32GB) + AMD EPYC 7543 32-Core Processor (120c) + 945GB",
   "model": "Qwen2.5-Coder-14B-Instruct",
   "model_size": null,
   "quant": "Q4_K_M",
   "runtime": "llama.cpp",
   "runtime_version": "0.7.0-dev (build 1, commit 1268eda)",
   "workload": "agent-trace",
   "suite_version": "suite-v1",
   "decode_tps": 128.23,
   "prefill_tps": 18881.93,
   "ttft_ms": 112.6,
   "prompt_tokens": 3268,
   "output_tokens": 840,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_vsype4pm1ko",
   "run_url": "https://llm-speed.com/r/r_vsype4pm1ko",
   "date": "2026-10-17",
   "hardware": "RTX 5090 (32GB) + AMD EPYC 7543 32-Core Processor (120c) + 945GB",
   "model": "Qwen3-Coder-30B-A3B-Instruct",
   "model_size": null,
   "quant": "Q4_K_M",
   "runtime": "llama.cpp",
   "runtime_version": "0.7.0-dev (build 1, commit 1268eda)",
   "workload": "concurrent-decode",
   "suite_version": "suite-v1",
   "decode_tps": 296.72,
   "prefill_tps": null,
   "ttft_ms": null,
   "prompt_tokens": 1024,
   "output_tokens": 256,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_vsype4pm1ko",
   "run_url": "https://llm-speed.com/r/r_vsype4pm1ko",
   "date": "2026-10-17",
   "hardware": "RTX 5090 (32GB) + AMD EPYC 7543 32-Core Processor (120c) + 945GB",
   "model": "Qwen3-Coder-30B-A3B-Instruct",
   "model_size": null,
   "quant": "Q4_K_M",
   "runtime": "llama.cpp",
   "runtime_version": "0.7.0-dev (build 1, commit 1268eda)",
   "workload": "chat-short",
   "suite_version": "suite-v1",
   "decode_tps": 304.9,
   "prefill_tps": 514.53,
   "ttft_ms": 213.8,
   "prompt_tokens": 110,
   "output_tokens": 256,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_vsype4pm1ko",
   "run_url": "https://llm-speed.com/r/r_vsype4pm1ko",
   "date": "2026-10-17",
   "hardware": "RTX 5090 (32GB) + AMD EPYC 7543 32-Core Processor (120c) + 945GB",
   "model": "Qwen3-Coder-30B-A3B-Instruct",
   "model_size": null,
   "quant": "Q4_K_M",
   "runtime": "llama.cpp",
   "runtime_version": "0.7.0-dev (build 1, commit 1268eda)",
   "workload": "chat-long",
   "suite_version": "suite-v1",
   "decode_tps": 266.16,
   "prefill_tps": 8353.17,
   "ttft_ms": 376.7,
   "prompt_tokens": 3147,
   "output_tokens": 557,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_vsype4pm1ko",
   "run_url": "https://llm-speed.com/r/r_vsype4pm1ko",
   "date": "2026-10-17",
   "hardware": "RTX 5090 (32GB) + AMD EPYC 7543 32-Core Processor (120c) + 945GB",
   "model": "Qwen3-Coder-30B-A3B-Instruct",
   "model_size": null,
   "quant": "Q4_K_M",
   "runtime": "llama.cpp",
   "runtime_version": "0.7.0-dev (build 1, commit 1268eda)",
   "workload": "agent-trace",
   "suite_version": "suite-v1",
   "decode_tps": 278.8,
   "prefill_tps": 23318.66,
   "ttft_ms": 96.6,
   "prompt_tokens": 3247,
   "output_tokens": 840,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_vhbqyt1f_xr",
   "run_url": "https://llm-speed.com/r/r_vhbqyt1f_xr",
   "date": "2026-10-17",
   "hardware": "RTX 5090 (32GB) + AMD EPYC 7543 32-Core Processor (120c) + 945GB",
   "model": "Qwen3-32B",
   "model_size": null,
   "quant": "Q4_K_M",
   "runtime": "llama.cpp",
   "runtime_version": "0.7.0-dev (build 1, commit 1268eda)",
   "workload": "concurrent-decode",
   "suite_version": "suite-v1",
   "decode_tps": 68.52,
   "prefill_tps": null,
   "ttft_ms": null,
   "prompt_tokens": 1024,
   "output_tokens": 256,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_vhbqyt1f_xr",
   "run_url": "https://llm-speed.com/r/r_vhbqyt1f_xr",
   "date": "2026-10-17",
   "hardware": "RTX 5090 (32GB) + AMD EPYC 7543 32-Core Processor (120c) + 945GB",
   "model": "Qwen3-32B",
   "model_size": null,
   "quant": "Q4_K_M",
   "runtime": "llama.cpp",
   "runtime_version": "0.7.0-dev (build 1, commit 1268eda)",
   "workload": "chat-short",
   "suite_version": "suite-v1",
   "decode_tps": 69.83,
   "prefill_tps": 1046.99,
   "ttft_ms": 105.1,
   "prompt_tokens": 110,
   "output_tokens": 256,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_vhbqyt1f_xr",
   "run_url": "https://llm-speed.com/r/r_vhbqyt1f_xr",
   "date": "2026-10-17",
   "hardware": "RTX 5090 (32GB) + AMD EPYC 7543 32-Core Processor (120c) + 945GB",
   "model": "Qwen3-32B",
   "model_size": null,
   "quant": "Q4_K_M",
   "runtime": "llama.cpp",
   "runtime_version": "0.7.0-dev (build 1, commit 1268eda)",
   "workload": "chat-long",
   "suite_version": "suite-v1",
   "decode_tps": 64.13,
   "prefill_tps": 3139.64,
   "ttft_ms": 1002.3,
   "prompt_tokens": 3147,
   "output_tokens": 1024,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_vhbqyt1f_xr",
   "run_url": "https://llm-speed.com/r/r_vhbqyt1f_xr",
   "date": "2026-10-17",
   "hardware": "RTX 5090 (32GB) + AMD EPYC 7543 32-Core Processor (120c) + 945GB",
   "model": "Qwen3-32B",
   "model_size": null,
   "quant": "Q4_K_M",
   "runtime": "llama.cpp",
   "runtime_version": "0.7.0-dev (build 1, commit 1268eda)",
   "workload": "agent-trace",
   "suite_version": "suite-v1",
   "decode_tps": 66.46,
   "prefill_tps": 8161.49,
   "ttft_ms": 252.6,
   "prompt_tokens": 3247,
   "output_tokens": 840,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_rkyf9wxzw9z",
   "run_url": "https://llm-speed.com/r/r_rkyf9wxzw9z",
   "date": "2026-10-17",
   "hardware": "RTX 3090 (24GB) + AMD EPYC 7282 16-Core Processor (16c) + 126GB",
   "model": "Qwen2.5-Coder-14B-Instruct",
   "model_size": null,
   "quant": "Q4_K_M",
   "runtime": "llama.cpp",
   "runtime_version": "0.7.0-dev (build 1, commit ea0c4c2)",
   "workload": "concurrent-decode",
   "suite_version": "suite-v1",
   "decode_tps": 77.11,
   "prefill_tps": null,
   "ttft_ms": null,
   "prompt_tokens": 1024,
   "output_tokens": 256,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_rkyf9wxzw9z",
   "run_url": "https://llm-speed.com/r/r_rkyf9wxzw9z",
   "date": "2026-10-17",
   "hardware": "RTX 3090 (24GB) + AMD EPYC 7282 16-Core Processor (16c) + 126GB",
   "model": "Qwen2.5-Coder-14B-Instruct",
   "model_size": null,
   "quant": "Q4_K_M",
   "runtime": "llama.cpp",
   "runtime_version": "0.7.0-dev (build 1, commit ea0c4c2)",
   "workload": "chat-short",
   "suite_version": "suite-v1",
   "decode_tps": 78.21,
   "prefill_tps": 1693.25,
   "ttft_ms": 77.4,
   "prompt_tokens": 131,
   "output_tokens": 256,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_rkyf9wxzw9z",
   "run_url": "https://llm-speed.com/r/r_rkyf9wxzw9z",
   "date": "2026-10-17",
   "hardware": "RTX 3090 (24GB) + AMD EPYC 7282 16-Core Processor (16c) + 126GB",
   "model": "Qwen2.5-Coder-14B-Instruct",
   "model_size": null,
   "quant": "Q4_K_M",
   "runtime": "llama.cpp",
   "runtime_version": "0.7.0-dev (build 1, commit ea0c4c2)",
   "workload": "chat-long",
   "suite_version": "suite-v1",
   "decode_tps": 74.31,
   "prefill_tps": 2549.2,
   "ttft_ms": 1242.7,
   "prompt_tokens": 3168,
   "output_tokens": 436,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_rkyf9wxzw9z",
   "run_url": "https://llm-speed.com/r/r_rkyf9wxzw9z",
   "date": "2026-10-17",
   "hardware": "RTX 3090 (24GB) + AMD EPYC 7282 16-Core Processor (16c) + 126GB",
   "model": "Qwen2.5-Coder-14B-Instruct",
   "model_size": null,
   "quant": "Q4_K_M",
   "runtime": "llama.cpp",
   "runtime_version": "0.7.0-dev (build 1, commit ea0c4c2)",
   "workload": "agent-trace",
   "suite_version": "suite-v1",
   "decode_tps": 75.86,
   "prefill_tps": 7427.04,
   "ttft_ms": 272.7,
   "prompt_tokens": 3268,
   "output_tokens": 840,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_qm1xyvjmx_r",
   "run_url": "https://llm-speed.com/r/r_qm1xyvjmx_r",
   "date": "2026-10-17",
   "hardware": "RTX 3090 (24GB) + AMD EPYC 7282 16-Core Processor (16c) + 126GB",
   "model": "Qwen3.6-27B",
   "model_size": null,
   "quant": "Q4_K_M",
   "runtime": "llama.cpp",
   "runtime_version": "0.7.0-dev (build 1, commit ea0c4c2)",
   "workload": "concurrent-decode",
   "suite_version": "suite-v1",
   "decode_tps": 42.63,
   "prefill_tps": null,
   "ttft_ms": null,
   "prompt_tokens": 1024,
   "output_tokens": 256,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_qm1xyvjmx_r",
   "run_url": "https://llm-speed.com/r/r_qm1xyvjmx_r",
   "date": "2026-10-17",
   "hardware": "RTX 3090 (24GB) + AMD EPYC 7282 16-Core Processor (16c) + 126GB",
   "model": "Qwen3.6-27B",
   "model_size": null,
   "quant": "Q4_K_M",
   "runtime": "llama.cpp",
   "runtime_version": "0.7.0-dev (build 1, commit ea0c4c2)",
   "workload": "chat-short",
   "suite_version": "suite-v1",
   "decode_tps": 42.73,
   "prefill_tps": 236.93,
   "ttft_ms": 472.7,
   "prompt_tokens": 112,
   "output_tokens": 256,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_qm1xyvjmx_r",
   "run_url": "https://llm-speed.com/r/r_qm1xyvjmx_r",
   "date": "2026-10-17",
   "hardware": "RTX 3090 (24GB) + AMD EPYC 7282 16-Core Processor (16c) + 126GB",
   "model": "Qwen3.6-27B",
   "model_size": null,
   "quant": "Q4_K_M",
   "runtime": "llama.cpp",
   "runtime_version": "0.7.0-dev (build 1, commit ea0c4c2)",
   "workload": "chat-long",
   "suite_version": "suite-v1",
   "decode_tps": 42.26,
   "prefill_tps": 1321.53,
   "ttft_ms": 2407.8,
   "prompt_tokens": 3182,
   "output_tokens": 1024,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_qm1xyvjmx_r",
   "run_url": "https://llm-speed.com/r/r_qm1xyvjmx_r",
   "date": "2026-10-17",
   "hardware": "RTX 3090 (24GB) + AMD EPYC 7282 16-Core Processor (16c) + 126GB",
   "model": "Qwen3.6-27B",
   "model_size": null,
   "quant": "Q4_K_M",
   "runtime": "llama.cpp",
   "runtime_version": "0.7.0-dev (build 1, commit ea0c4c2)",
   "workload": "agent-trace",
   "suite_version": "suite-v1",
   "decode_tps": 42.07,
   "prefill_tps": 2369.6,
   "ttft_ms": 881.0,
   "prompt_tokens": 3454,
   "output_tokens": 840,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_pob-co51bd-",
   "run_url": "https://llm-speed.com/r/r_pob-co51bd-",
   "date": "2026-10-17",
   "hardware": "RTX 5090 (32GB) + AMD EPYC 7543 32-Core Processor (120c) + 945GB",
   "model": "Qwen3.6-27B",
   "model_size": null,
   "quant": "Q4_K_M",
   "runtime": "llama.cpp",
   "runtime_version": "0.7.0-dev (build 1, commit 1268eda)",
   "workload": "concurrent-decode",
   "suite_version": "suite-v1",
   "decode_tps": 75.13,
   "prefill_tps": null,
   "ttft_ms": null,
   "prompt_tokens": 1024,
   "output_tokens": 256,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_pob-co51bd-",
   "run_url": "https://llm-speed.com/r/r_pob-co51bd-",
   "date": "2026-10-17",
   "hardware": "RTX 5090 (32GB) + AMD EPYC 7543 32-Core Processor (120c) + 945GB",
   "model": "Qwen3.6-27B",
   "model_size": null,
   "quant": "Q4_K_M",
   "runtime": "llama.cpp",
   "runtime_version": "0.7.0-dev (build 1, commit 1268eda)",
   "workload": "chat-short",
   "suite_version": "suite-v1",
   "decode_tps": 75.09,
   "prefill_tps": 302.22,
   "ttft_ms": 370.6,
   "prompt_tokens": 112,
   "output_tokens": 256,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_pob-co51bd-",
   "run_url": "https://llm-speed.com/r/r_pob-co51bd-",
   "date": "2026-10-17",
   "hardware": "RTX 5090 (32GB) + AMD EPYC 7543 32-Core Processor (120c) + 945GB",
   "model": "Qwen3.6-27B",
   "model_size": null,
   "quant": "Q4_K_M",
   "runtime": "llama.cpp",
   "runtime_version": "0.7.0-dev (build 1, commit 1268eda)",
   "workload": "chat-long",
   "suite_version": "suite-v1",
   "decode_tps": 74.55,
   "prefill_tps": 2644.79,
   "ttft_ms": 1203.1,
   "prompt_tokens": 3182,
   "output_tokens": 1024,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_pob-co51bd-",
   "run_url": "https://llm-speed.com/r/r_pob-co51bd-",
   "date": "2026-10-17",
   "hardware": "RTX 5090 (32GB) + AMD EPYC 7543 32-Core Processor (120c) + 945GB",
   "model": "Qwen3.6-27B",
   "model_size": null,
   "quant": "Q4_K_M",
   "runtime": "llama.cpp",
   "runtime_version": "0.7.0-dev (build 1, commit 1268eda)",
   "workload": "agent-trace",
   "suite_version": "suite-v1",
   "decode_tps": 74.13,
   "prefill_tps": 4464.52,
   "ttft_ms": 485.3,
   "prompt_tokens": 3454,
   "output_tokens": 840,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_nakw1nq6clw",
   "run_url": "https://llm-speed.com/r/r_nakw1nq6clw",
   "date": "2026-10-17",
   "hardware": "RTX 3090 (24GB) + AMD EPYC 7282 16-Core Processor (16c) + 126GB",
   "model": "gpt-oss-20b",
   "model_size": null,
   "quant": "Q4_K_M",
   "runtime": "llama.cpp",
   "runtime_version": "0.7.0-dev (build 1, commit ea0c4c2)",
   "workload": "concurrent-decode",
   "suite_version": "suite-v1",
   "decode_tps": 205.11,
   "prefill_tps": null,
   "ttft_ms": null,
   "prompt_tokens": 1024,
   "output_tokens": 256,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_nakw1nq6clw",
   "run_url": "https://llm-speed.com/r/r_nakw1nq6clw",
   "date": "2026-10-17",
   "hardware": "RTX 3090 (24GB) + AMD EPYC 7282 16-Core Processor (16c) + 126GB",
   "model": "gpt-oss-20b",
   "model_size": null,
   "quant": "Q4_K_M",
   "runtime": "llama.cpp",
   "runtime_version": "0.7.0-dev (build 1, commit ea0c4c2)",
   "workload": "chat-short",
   "suite_version": "suite-v1",
   "decode_tps": 207.2,
   "prefill_tps": 658.34,
   "ttft_ms": 252.1,
   "prompt_tokens": 166,
   "output_tokens": 256,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_nakw1nq6clw",
   "run_url": "https://llm-speed.com/r/r_nakw1nq6clw",
   "date": "2026-10-17",
   "hardware": "RTX 3090 (24GB) + AMD EPYC 7282 16-Core Processor (16c) + 126GB",
   "model": "gpt-oss-20b",
   "model_size": null,
   "quant": "Q4_K_M",
   "runtime": "llama.cpp",
   "runtime_version": "0.7.0-dev (build 1, commit ea0c4c2)",
   "workload": "chat-long",
   "suite_version": "suite-v1",
   "decode_tps": 199.82,
   "prefill_tps": 4606.27,
   "ttft_ms": 692.1,
   "prompt_tokens": 3188,
   "output_tokens": 998,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_nakw1nq6clw",
   "run_url": "https://llm-speed.com/r/r_nakw1nq6clw",
   "date": "2026-10-17",
   "hardware": "RTX 3090 (24GB) + AMD EPYC 7282 16-Core Processor (16c) + 126GB",
   "model": "gpt-oss-20b",
   "model_size": null,
   "quant": "Q4_K_M",
   "runtime": "llama.cpp",
   "runtime_version": "0.7.0-dev (build 1, commit ea0c4c2)",
   "workload": "agent-trace",
   "suite_version": "suite-v1",
   "decode_tps": 202.17,
   "prefill_tps": 9892.74,
   "ttft_ms": 224.0,
   "prompt_tokens": 3287,
   "output_tokens": 840,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_n93ukk0mkmm",
   "run_url": "https://llm-speed.com/r/r_n93ukk0mkmm",
   "date": "2026-10-17",
   "hardware": "2x RTX 3090 (24GB) + AMD EPYC 7H12 64-Core Processor (128c) + 1008GB",
   "model": "Llama-3.3-70B-Instruct",
   "model_size": null,
   "quant": "Q4_K_M",
   "runtime": "llama.cpp",
   "runtime_version": "0.7.0-dev (build 1, commit f74e2fa)",
   "workload": "concurrent-decode",
   "suite_version": "suite-v1",
   "decode_tps": 19.0,
   "prefill_tps": null,
   "ttft_ms": null,
   "prompt_tokens": 1024,
   "output_tokens": 256,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_n93ukk0mkmm",
   "run_url": "https://llm-speed.com/r/r_n93ukk0mkmm",
   "date": "2026-10-17",
   "hardware": "2x RTX 3090 (24GB) + AMD EPYC 7H12 64-Core Processor (128c) + 1008GB",
   "model": "Llama-3.3-70B-Instruct",
   "model_size": null,
   "quant": "Q4_K_M",
   "runtime": "llama.cpp",
   "runtime_version": "0.7.0-dev (build 1, commit f74e2fa)",
   "workload": "chat-short",
   "suite_version": "suite-v1",
   "decode_tps": 19.14,
   "prefill_tps": 415.39,
   "ttft_ms": 325.0,
   "prompt_tokens": 135,
   "output_tokens": 256,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_n93ukk0mkmm",
   "run_url": "https://llm-speed.com/r/r_n93ukk0mkmm",
   "date": "2026-10-17",
   "hardware": "2x RTX 3090 (24GB) + AMD EPYC 7H12 64-Core Processor (128c) + 1008GB",
   "model": "Llama-3.3-70B-Instruct",
   "model_size": null,
   "quant": "Q4_K_M",
   "runtime": "llama.cpp",
   "runtime_version": "0.7.0-dev (build 1, commit f74e2fa)",
   "workload": "chat-long",
   "suite_version": "suite-v1",
   "decode_tps": 18.69,
   "prefill_tps": 939.28,
   "ttft_ms": 3366.4,
   "prompt_tokens": 3162,
   "output_tokens": 452,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_n93ukk0mkmm",
   "run_url": "https://llm-speed.com/r/r_n93ukk0mkmm",
   "date": "2026-10-17",
   "hardware": "2x RTX 3090 (24GB) + AMD EPYC 7H12 64-Core Processor (128c) + 1008GB",
   "model": "Llama-3.3-70B-Instruct",
   "model_size": null,
   "quant": "Q4_K_M",
   "runtime": "llama.cpp",
   "runtime_version": "0.7.0-dev (build 1, commit f74e2fa)",
   "workload": "agent-trace",
   "suite_version": "suite-v1",
   "decode_tps": 18.85,
   "prefill_tps": 2028.77,
   "ttft_ms": 1037.0,
   "prompt_tokens": 3235,
   "output_tokens": 840,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_km62sybxga6",
   "run_url": "https://llm-speed.com/r/r_km62sybxga6",
   "date": "2026-10-17",
   "hardware": "RTX 3090 (24GB) + AMD EPYC 7282 16-Core Processor (16c) + 126GB",
   "model": "Qwen3-32B",
   "model_size": null,
   "quant": "Q4_K_M",
   "runtime": "llama.cpp",
   "runtime_version": "0.7.0-dev (build 1, commit ea0c4c2)",
   "workload": "concurrent-decode",
   "suite_version": "suite-v1",
   "decode_tps": 38.05,
   "prefill_tps": null,
   "ttft_ms": null,
   "prompt_tokens": 1024,
   "output_tokens": 256,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_km62sybxga6",
   "run_url": "https://llm-speed.com/r/r_km62sybxga6",
   "date": "2026-10-17",
   "hardware": "RTX 3090 (24GB) + AMD EPYC 7282 16-Core Processor (16c) + 126GB",
   "model": "Qwen3-32B",
   "model_size": null,
   "quant": "Q4_K_M",
   "runtime": "llama.cpp",
   "runtime_version": "0.7.0-dev (build 1, commit ea0c4c2)",
   "workload": "chat-short",
   "suite_version": "suite-v1",
   "decode_tps": 38.45,
   "prefill_tps": 589.55,
   "ttft_ms": 186.6,
   "prompt_tokens": 110,
   "output_tokens": 256,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_km62sybxga6",
   "run_url": "https://llm-speed.com/r/r_km62sybxga6",
   "date": "2026-10-17",
   "hardware": "RTX 3090 (24GB) + AMD EPYC 7282 16-Core Processor (16c) + 126GB",
   "model": "Qwen3-32B",
   "model_size": null,
   "quant": "Q4_K_M",
   "runtime": "llama.cpp",
   "runtime_version": "0.7.0-dev (build 1, commit ea0c4c2)",
   "workload": "chat-long",
   "suite_version": "suite-v1",
   "decode_tps": 36.94,
   "prefill_tps": 1218.15,
   "ttft_ms": 2583.4,
   "prompt_tokens": 3147,
   "output_tokens": 1024,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_km62sybxga6",
   "run_url": "https://llm-speed.com/r/r_km62sybxga6",
   "date": "2026-10-17",
   "hardware": "RTX 3090 (24GB) + AMD EPYC 7282 16-Core Processor (16c) + 126GB",
   "model": "Qwen3-32B",
   "model_size": null,
   "quant": "Q4_K_M",
   "runtime": "llama.cpp",
   "runtime_version": "0.7.0-dev (build 1, commit ea0c4c2)",
   "workload": "agent-trace",
   "suite_version": "suite-v1",
   "decode_tps": 37.57,
   "prefill_tps": 3287.65,
   "ttft_ms": 623.9,
   "prompt_tokens": 3247,
   "output_tokens": 840,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_dwp5fgdapmw",
   "run_url": "https://llm-speed.com/r/r_dwp5fgdapmw",
   "date": "2026-10-17",
   "hardware": "RTX 5090 (32GB) + AMD EPYC 7543 32-Core Processor (120c) + 945GB",
   "model": "gpt-oss-20b",
   "model_size": null,
   "quant": "Q4_K_M",
   "runtime": "llama.cpp",
   "runtime_version": "0.7.0-dev (build 1, commit 1268eda)",
   "workload": "concurrent-decode",
   "suite_version": "suite-v1",
   "decode_tps": 322.46,
   "prefill_tps": null,
   "ttft_ms": null,
   "prompt_tokens": 1024,
   "output_tokens": 256,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_dwp5fgdapmw",
   "run_url": "https://llm-speed.com/r/r_dwp5fgdapmw",
   "date": "2026-10-17",
   "hardware": "RTX 5090 (32GB) + AMD EPYC 7543 32-Core Processor (120c) + 945GB",
   "model": "gpt-oss-20b",
   "model_size": null,
   "quant": "Q4_K_M",
   "runtime": "llama.cpp",
   "runtime_version": "0.7.0-dev (build 1, commit 1268eda)",
   "workload": "chat-short",
   "suite_version": "suite-v1",
   "decode_tps": 324.2,
   "prefill_tps": 586.77,
   "ttft_ms": 282.9,
   "prompt_tokens": 166,
   "output_tokens": 256,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_dwp5fgdapmw",
   "run_url": "https://llm-speed.com/r/r_dwp5fgdapmw",
   "date": "2026-10-17",
   "hardware": "RTX 5090 (32GB) + AMD EPYC 7543 32-Core Processor (120c) + 945GB",
   "model": "gpt-oss-20b",
   "model_size": null,
   "quant": "Q4_K_M",
   "runtime": "llama.cpp",
   "runtime_version": "0.7.0-dev (build 1, commit 1268eda)",
   "workload": "chat-long",
   "suite_version": "suite-v1",
   "decode_tps": 310.98,
   "prefill_tps": 12638.44,
   "ttft_ms": 252.2,
   "prompt_tokens": 3188,
   "output_tokens": 1024,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_dwp5fgdapmw",
   "run_url": "https://llm-speed.com/r/r_dwp5fgdapmw",
   "date": "2026-10-17",
   "hardware": "RTX 5090 (32GB) + AMD EPYC 7543 32-Core Processor (120c) + 945GB",
   "model": "gpt-oss-20b",
   "model_size": null,
   "quant": "Q4_K_M",
   "runtime": "llama.cpp",
   "runtime_version": "0.7.0-dev (build 1, commit 1268eda)",
   "workload": "agent-trace",
   "suite_version": "suite-v1",
   "decode_tps": 319.64,
   "prefill_tps": 23886.95,
   "ttft_ms": 89.2,
   "prompt_tokens": 3287,
   "output_tokens": 840,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_0dh1kzbv4h0",
   "run_url": "https://llm-speed.com/r/r_0dh1kzbv4h0",
   "date": "2026-10-17",
   "hardware": "RTX 3090 (24GB) + AMD EPYC 7282 16-Core Processor (16c) + 126GB",
   "model": "Qwen3-Coder-30B-A3B-Instruct",
   "model_size": null,
   "quant": "Q4_K_M",
   "runtime": "llama.cpp",
   "runtime_version": "0.7.0-dev (build 1, commit ea0c4c2)",
   "workload": "concurrent-decode",
   "suite_version": "suite-v1",
   "decode_tps": 198.04,
   "prefill_tps": null,
   "ttft_ms": null,
   "prompt_tokens": 1024,
   "output_tokens": 256,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_0dh1kzbv4h0",
   "run_url": "https://llm-speed.com/r/r_0dh1kzbv4h0",
   "date": "2026-10-17",
   "hardware": "RTX 3090 (24GB) + AMD EPYC 7282 16-Core Processor (16c) + 126GB",
   "model": "Qwen3-Coder-30B-A3B-Instruct",
   "model_size": null,
   "quant": "Q4_K_M",
   "runtime": "llama.cpp",
   "runtime_version": "0.7.0-dev (build 1, commit ea0c4c2)",
   "workload": "chat-short",
   "suite_version": "suite-v1",
   "decode_tps": 199.73,
   "prefill_tps": 612.55,
   "ttft_ms": 179.6,
   "prompt_tokens": 110,
   "output_tokens": 256,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_0dh1kzbv4h0",
   "run_url": "https://llm-speed.com/r/r_0dh1kzbv4h0",
   "date": "2026-10-17",
   "hardware": "RTX 3090 (24GB) + AMD EPYC 7282 16-Core Processor (16c) + 126GB",
   "model": "Qwen3-Coder-30B-A3B-Instruct",
   "model_size": null,
   "quant": "Q4_K_M",
   "runtime": "llama.cpp",
   "runtime_version": "0.7.0-dev (build 1, commit ea0c4c2)",
   "workload": "chat-long",
   "suite_version": "suite-v1",
   "decode_tps": 186.98,
   "prefill_tps": 3971.3,
   "ttft_ms": 792.4,
   "prompt_tokens": 3147,
   "output_tokens": 580,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_0dh1kzbv4h0",
   "run_url": "https://llm-speed.com/r/r_0dh1kzbv4h0",
   "date": "2026-10-17",
   "hardware": "RTX 3090 (24GB) + AMD EPYC 7282 16-Core Processor (16c) + 126GB",
   "model": "Qwen3-Coder-30B-A3B-Instruct",
   "model_size": null,
   "quant": "Q4_K_M",
   "runtime": "llama.cpp",
   "runtime_version": "0.7.0-dev (build 1, commit ea0c4c2)",
   "workload": "agent-trace",
   "suite_version": "suite-v1",
   "decode_tps": 190.91,
   "prefill_tps": 10182.2,
   "ttft_ms": 214.3,
   "prompt_tokens": 3247,
   "output_tokens": 840,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_ted0u8uyre0",
   "run_url": "https://llm-speed.com/r/r_ted0u8uyre0",
   "date": "2026-10-16",
   "hardware": "RTX 4090 (24GB) + AMD EPYC 75F3 32-Core Processor (64c) + 504GB",
   "model": "glm-4.7-flash",
   "model_size": "29.9B",
   "quant": "Q4_K_M",
   "runtime": "ollama",
   "runtime_version": "0.31.1",
   "workload": "concurrent-decode",
   "suite_version": "suite-v1",
   "decode_tps": 155.23,
   "prefill_tps": null,
   "ttft_ms": null,
   "prompt_tokens": 1024,
   "output_tokens": 256,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_ted0u8uyre0",
   "run_url": "https://llm-speed.com/r/r_ted0u8uyre0",
   "date": "2026-10-16",
   "hardware": "RTX 4090 (24GB) + AMD EPYC 75F3 32-Core Processor (64c) + 504GB",
   "model": "glm-4.7-flash",
   "model_size": "29.9B",
   "quant": "Q4_K_M",
   "runtime": "ollama",
   "runtime_version": "0.31.1",
   "workload": "chat-short",
   "suite_version": "suite-v1",
   "decode_tps": 164.94,
   "prefill_tps": 3.74,
   "ttft_ms": 28105.7,
   "prompt_tokens": 105,
   "output_tokens": 256,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_ted0u8uyre0",
   "run_url": "https://llm-speed.com/r/r_ted0u8uyre0",
   "date": "2026-10-16",
   "hardware": "RTX 4090 (24GB) + AMD EPYC 75F3 32-Core Processor (64c) + 504GB",
   "model": "glm-4.7-flash",
   "model_size": "29.9B",
   "quant": "Q4_K_M",
   "runtime": "ollama",
   "runtime_version": "0.31.1",
   "workload": "chat-long",
   "suite_version": "suite-v1",
   "decode_tps": 146.1,
   "prefill_tps": 3617.69,
   "ttft_ms": 865.7,
   "prompt_tokens": 3132,
   "output_tokens": 1024,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_ted0u8uyre0",
   "run_url": "https://llm-speed.com/r/r_ted0u8uyre0",
   "date": "2026-10-16",
   "hardware": "RTX 4090 (24GB) + AMD EPYC 75F3 32-Core Processor (64c) + 504GB",
   "model": "glm-4.7-flash",
   "model_size": "29.9B",
   "quant": "Q4_K_M",
   "runtime": "ollama",
   "runtime_version": "0.31.1",
   "workload": "agent-trace",
   "suite_version": "suite-v1",
   "decode_tps": 155.54,
   "prefill_tps": 3224.64,
   "ttft_ms": 641.5,
   "prompt_tokens": 3145,
   "output_tokens": 840,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_m6j1zxwvp-f",
   "run_url": "https://llm-speed.com/r/r_m6j1zxwvp-f",
   "date": "2026-10-16",
   "hardware": "RTX 4090 (24GB) + AMD EPYC 7642 48-Core Processor (48c) + 252GB",
   "model": "Qwen3.6-35B-A3B",
   "model_size": null,
   "quant": "Q4_K_M",
   "runtime": "llama.cpp",
   "runtime_version": "8931 (9725a313b)",
   "workload": "concurrent-decode",
   "suite_version": "suite-v1",
   "decode_tps": 163.46,
   "prefill_tps": null,
   "ttft_ms": null,
   "prompt_tokens": 1024,
   "output_tokens": 256,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_m6j1zxwvp-f",
   "run_url": "https://llm-speed.com/r/r_m6j1zxwvp-f",
   "date": "2026-10-16",
   "hardware": "RTX 4090 (24GB) + AMD EPYC 7642 48-Core Processor (48c) + 252GB",
   "model": "Qwen3.6-35B-A3B",
   "model_size": null,
   "quant": "Q4_K_M",
   "runtime": "llama.cpp",
   "runtime_version": "8931 (9725a313b)",
   "workload": "chat-short",
   "suite_version": "suite-v1",
   "decode_tps": 160.59,
   "prefill_tps": 464.33,
   "ttft_ms": 241.2,
   "prompt_tokens": 112,
   "output_tokens": 256,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_m6j1zxwvp-f",
   "run_url": "https://llm-speed.com/r/r_m6j1zxwvp-f",
   "date": "2026-10-16",
   "hardware": "RTX 4090 (24GB) + AMD EPYC 7642 48-Core Processor (48c) + 252GB",
   "model": "Qwen3.6-35B-A3B",
   "model_size": null,
   "quant": "Q4_K_M",
   "runtime": "llama.cpp",
   "runtime_version": "8931 (9725a313b)",
   "workload": "chat-long",
   "suite_version": "suite-v1",
   "decode_tps": 160.39,
   "prefill_tps": 4588.77,
   "ttft_ms": 693.4,
   "prompt_tokens": 3182,
   "output_tokens": 1024,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_m6j1zxwvp-f",
   "run_url": "https://llm-speed.com/r/r_m6j1zxwvp-f",
   "date": "2026-10-16",
   "hardware": "RTX 4090 (24GB) + AMD EPYC 7642 48-Core Processor (48c) + 252GB",
   "model": "Qwen3.6-35B-A3B",
   "model_size": null,
   "quant": "Q4_K_M",
   "runtime": "llama.cpp",
   "runtime_version": "8931 (9725a313b)",
   "workload": "agent-trace",
   "suite_version": "suite-v1",
   "decode_tps": 162.69,
   "prefill_tps": 6547.51,
   "ttft_ms": 330.0,
   "prompt_tokens": 3454,
   "output_tokens": 840,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_kalwcyab-86",
   "run_url": "https://llm-speed.com/r/r_kalwcyab-86",
   "date": "2026-10-16",
   "hardware": "RTX 3090 (24GB) + AMD Ryzen Threadripper PRO 5975WX 32-Cores (32c) + 504GB",
   "model": "glm-4.7-flash",
   "model_size": "29.9B",
   "quant": "Q4_K_M",
   "runtime": "ollama",
   "runtime_version": "0.31.1",
   "workload": "concurrent-decode",
   "suite_version": "suite-v1",
   "decode_tps": 133.56,
   "prefill_tps": null,
   "ttft_ms": null,
   "prompt_tokens": 1024,
   "output_tokens": 256,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_kalwcyab-86",
   "run_url": "https://llm-speed.com/r/r_kalwcyab-86",
   "date": "2026-10-16",
   "hardware": "RTX 3090 (24GB) + AMD Ryzen Threadripper PRO 5975WX 32-Cores (32c) + 504GB",
   "model": "glm-4.7-flash",
   "model_size": "29.9B",
   "quant": "Q4_K_M",
   "runtime": "ollama",
   "runtime_version": "0.31.1",
   "workload": "chat-short",
   "suite_version": "suite-v1",
   "decode_tps": 134.62,
   "prefill_tps": 4.32,
   "ttft_ms": 24308.6,
   "prompt_tokens": 105,
   "output_tokens": 256,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_kalwcyab-86",
   "run_url": "https://llm-speed.com/r/r_kalwcyab-86",
   "date": "2026-10-16",
   "hardware": "RTX 3090 (24GB) + AMD Ryzen Threadripper PRO 5975WX 32-Cores (32c) + 504GB",
   "model": "glm-4.7-flash",
   "model_size": "29.9B",
   "quant": "Q4_K_M",
   "runtime": "ollama",
   "runtime_version": "0.31.1",
   "workload": "chat-long",
   "suite_version": "suite-v1",
   "decode_tps": 116.07,
   "prefill_tps": 2090.07,
   "ttft_ms": 1498.5,
   "prompt_tokens": 3132,
   "output_tokens": 1024,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_kalwcyab-86",
   "run_url": "https://llm-speed.com/r/r_kalwcyab-86",
   "date": "2026-10-16",
   "hardware": "RTX 3090 (24GB) + AMD Ryzen Threadripper PRO 5975WX 32-Cores (32c) + 504GB",
   "model": "glm-4.7-flash",
   "model_size": "29.9B",
   "quant": "Q4_K_M",
   "runtime": "ollama",
   "runtime_version": "0.31.1",
   "workload": "agent-trace",
   "suite_version": "suite-v1",
   "decode_tps": 125.17,
   "prefill_tps": 2901.43,
   "ttft_ms": 614.5,
   "prompt_tokens": 3145,
   "output_tokens": 840,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_jfh0qkngw76",
   "run_url": "https://llm-speed.com/r/r_jfh0qkngw76",
   "date": "2026-10-16",
   "hardware": "RTX 4090 (24GB) + AMD EPYC 7642 48-Core Processor (48c) + 252GB",
   "model": "gemma-4-12b-it-qat",
   "model_size": null,
   "quant": "Q4_0",
   "runtime": "llama.cpp",
   "runtime_version": "8931 (9725a313b)",
   "workload": "concurrent-decode",
   "suite_version": "suite-v1",
   "decode_tps": 95.49,
   "prefill_tps": null,
   "ttft_ms": null,
   "prompt_tokens": 1024,
   "output_tokens": 256,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_jfh0qkngw76",
   "run_url": "https://llm-speed.com/r/r_jfh0qkngw76",
   "date": "2026-10-16",
   "hardware": "RTX 4090 (24GB) + AMD EPYC 7642 48-Core Processor (48c) + 252GB",
   "model": "gemma-4-12b-it-qat",
   "model_size": null,
   "quant": "Q4_0",
   "runtime": "llama.cpp",
   "runtime_version": "8931 (9725a313b)",
   "workload": "chat-short",
   "suite_version": "suite-v1",
   "decode_tps": 97.72,
   "prefill_tps": 1109.05,
   "ttft_ms": 111.8,
   "prompt_tokens": 124,
   "output_tokens": 256,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_jfh0qkngw76",
   "run_url": "https://llm-speed.com/r/r_jfh0qkngw76",
   "date": "2026-10-16",
   "hardware": "RTX 4090 (24GB) + AMD EPYC 7642 48-Core Processor (48c) + 252GB",
   "model": "gemma-4-12b-it-qat",
   "model_size": null,
   "quant": "Q4_0",
   "runtime": "llama.cpp",
   "runtime_version": "8931 (9725a313b)",
   "workload": "chat-long",
   "suite_version": "suite-v1",
   "decode_tps": 86.3,
   "prefill_tps": 1742.99,
   "ttft_ms": 1830.8,
   "prompt_tokens": 3191,
   "output_tokens": 1024,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_jfh0qkngw76",
   "run_url": "https://llm-speed.com/r/r_jfh0qkngw76",
   "date": "2026-10-16",
   "hardware": "RTX 4090 (24GB) + AMD EPYC 7642 48-Core Processor (48c) + 252GB",
   "model": "gemma-4-12b-it-qat",
   "model_size": null,
   "quant": "Q4_0",
   "runtime": "llama.cpp",
   "runtime_version": "8931 (9725a313b)",
   "workload": "agent-trace",
   "suite_version": "suite-v1",
   "decode_tps": 89.1,
   "prefill_tps": 2305.05,
   "ttft_ms": 983.3,
   "prompt_tokens": 3939,
   "output_tokens": 840,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_fsd9qjmmna0",
   "run_url": "https://llm-speed.com/r/r_fsd9qjmmna0",
   "date": "2026-10-16",
   "hardware": "3x A100 (80GB) + AMD EPYC 7742 64-Core Processor (128c) + 1008GB",
   "model": "Muse-Spark-Open-816B-A42B-UD-00001-of-00005",
   "model_size": null,
   "quant": "IQ1_S",
   "runtime": "llama.cpp",
   "runtime_version": "0.7.0-dev (build 11675, commit 2851b3bcb)",
   "workload": "concurrent-decode",
   "suite_version": "suite-v1",
   "decode_tps": 26.3,
   "prefill_tps": null,
   "ttft_ms": null,
   "prompt_tokens": 1024,
   "output_tokens": 256,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_fsd9qjmmna0",
   "run_url": "https://llm-speed.com/r/r_fsd9qjmmna0",
   "date": "2026-10-16",
   "hardware": "3x A100 (80GB) + AMD EPYC 7742 64-Core Processor (128c) + 1008GB",
   "model": "Muse-Spark-Open-816B-A42B-UD-00001-of-00005",
   "model_size": null,
   "quant": "IQ1_S",
   "runtime": "llama.cpp",
   "runtime_version": "0.7.0-dev (build 11675, commit 2851b3bcb)",
   "workload": "chat-short",
   "suite_version": "suite-v1",
   "decode_tps": 25.2,
   "prefill_tps": 88.57,
   "ttft_ms": 2732.2,
   "prompt_tokens": 242,
   "output_tokens": 256,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_fsd9qjmmna0",
   "run_url": "https://llm-speed.com/r/r_fsd9qjmmna0",
   "date": "2026-10-16",
   "hardware": "3x A100 (80GB) + AMD EPYC 7742 64-Core Processor (128c) + 1008GB",
   "model": "Muse-Spark-Open-816B-A42B-UD-00001-of-00005",
   "model_size": null,
   "quant": "IQ1_S",
   "runtime": "llama.cpp",
   "runtime_version": "0.7.0-dev (build 11675, commit 2851b3bcb)",
   "workload": "chat-long",
   "suite_version": "suite-v1",
   "decode_tps": 25.5,
   "prefill_tps": 288.68,
   "ttft_ms": 11178.5,
   "prompt_tokens": 3227,
   "output_tokens": 1024,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_fsd9qjmmna0",
   "run_url": "https://llm-speed.com/r/r_fsd9qjmmna0",
   "date": "2026-10-16",
   "hardware": "3x A100 (80GB) + AMD EPYC 7742 64-Core Processor (128c) + 1008GB",
   "model": "Muse-Spark-Open-816B-A42B-UD-00001-of-00005",
   "model_size": null,
   "quant": "IQ1_S",
   "runtime": "llama.cpp",
   "runtime_version": "0.7.0-dev (build 11675, commit 2851b3bcb)",
   "workload": "agent-trace",
   "suite_version": "suite-v1",
   "decode_tps": 22.79,
   "prefill_tps": 570.75,
   "ttft_ms": 3895.5,
   "prompt_tokens": 3350,
   "output_tokens": 840,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_3gnxkzhhmyj",
   "run_url": "https://llm-speed.com/r/r_3gnxkzhhmyj",
   "date": "2026-10-16",
   "hardware": "RTX 3090 (24GB) + AMD Ryzen Threadripper PRO 5975WX 32-Cores (32c) + 504GB",
   "model": "gemma3",
   "model_size": "11.8B",
   "quant": "Q4_K_M",
   "runtime": "ollama",
   "runtime_version": "0.31.1",
   "workload": "concurrent-decode",
   "suite_version": "suite-v1",
   "decode_tps": 76.01,
   "prefill_tps": null,
   "ttft_ms": null,
   "prompt_tokens": 1024,
   "output_tokens": 256,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_3gnxkzhhmyj",
   "run_url": "https://llm-speed.com/r/r_3gnxkzhhmyj",
   "date": "2026-10-16",
   "hardware": "RTX 3090 (24GB) + AMD Ryzen Threadripper PRO 5975WX 32-Cores (32c) + 504GB",
   "model": "gemma3",
   "model_size": "11.8B",
   "quant": "Q4_K_M",
   "runtime": "ollama",
   "runtime_version": "0.31.1",
   "workload": "chat-short",
   "suite_version": "suite-v1",
   "decode_tps": 78.65,
   "prefill_tps": 26.02,
   "ttft_ms": 4534.1,
   "prompt_tokens": 118,
   "output_tokens": 256,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_3gnxkzhhmyj",
   "run_url": "https://llm-speed.com/r/r_3gnxkzhhmyj",
   "date": "2026-10-16",
   "hardware": "RTX 3090 (24GB) + AMD Ryzen Threadripper PRO 5975WX 32-Cores (32c) + 504GB",
   "model": "gemma3",
   "model_size": "11.8B",
   "quant": "Q4_K_M",
   "runtime": "ollama",
   "runtime_version": "0.31.1",
   "workload": "chat-long",
   "suite_version": "suite-v1",
   "decode_tps": 74.24,
   "prefill_tps": 1648.09,
   "ttft_ms": 1932.5,
   "prompt_tokens": 3185,
   "output_tokens": 751,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_3gnxkzhhmyj",
   "run_url": "https://llm-speed.com/r/r_3gnxkzhhmyj",
   "date": "2026-10-16",
   "hardware": "RTX 3090 (24GB) + AMD Ryzen Threadripper PRO 5975WX 32-Cores (32c) + 504GB",
   "model": "gemma3",
   "model_size": "11.8B",
   "quant": "Q4_K_M",
   "runtime": "ollama",
   "runtime_version": "0.31.1",
   "workload": "agent-trace",
   "suite_version": "suite-v1",
   "decode_tps": 74.48,
   "prefill_tps": 2887.72,
   "ttft_ms": 807.8,
   "prompt_tokens": 3934,
   "output_tokens": 840,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_19f64yux97o",
   "run_url": "https://llm-speed.com/r/r_19f64yux97o",
   "date": "2026-10-16",
   "hardware": "A40 (45GB) + Intel(R) Xeon(R) Gold 6342 CPU @ 2.80GHz (48c) + 504GB",
   "model": "Llama-3.3-70B-Instruct",
   "model_size": null,
   "quant": "Q4_K_M",
   "runtime": "llama.cpp",
   "runtime_version": "0.7.0-dev (build 1, commit ceefd16)",
   "workload": "concurrent-decode",
   "suite_version": "suite-v1",
   "decode_tps": 12.85,
   "prefill_tps": null,
   "ttft_ms": null,
   "prompt_tokens": 1024,
   "output_tokens": 256,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_19f64yux97o",
   "run_url": "https://llm-speed.com/r/r_19f64yux97o",
   "date": "2026-10-16",
   "hardware": "A40 (45GB) + Intel(R) Xeon(R) Gold 6342 CPU @ 2.80GHz (48c) + 504GB",
   "model": "Llama-3.3-70B-Instruct",
   "model_size": null,
   "quant": "Q4_K_M",
   "runtime": "llama.cpp",
   "runtime_version": "0.7.0-dev (build 1, commit ceefd16)",
   "workload": "chat-short",
   "suite_version": "suite-v1",
   "decode_tps": 12.99,
   "prefill_tps": 400.98,
   "ttft_ms": 336.7,
   "prompt_tokens": 135,
   "output_tokens": 256,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_19f64yux97o",
   "run_url": "https://llm-speed.com/r/r_19f64yux97o",
   "date": "2026-10-16",
   "hardware": "A40 (45GB) + Intel(R) Xeon(R) Gold 6342 CPU @ 2.80GHz (48c) + 504GB",
   "model": "Llama-3.3-70B-Instruct",
   "model_size": null,
   "quant": "Q4_K_M",
   "runtime": "llama.cpp",
   "runtime_version": "0.7.0-dev (build 1, commit ceefd16)",
   "workload": "chat-long",
   "suite_version": "suite-v1",
   "decode_tps": 12.67,
   "prefill_tps": 551.39,
   "ttft_ms": 5734.6,
   "prompt_tokens": 3162,
   "output_tokens": 458,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_19f64yux97o",
   "run_url": "https://llm-speed.com/r/r_19f64yux97o",
   "date": "2026-10-16",
   "hardware": "A40 (45GB) + Intel(R) Xeon(R) Gold 6342 CPU @ 2.80GHz (48c) + 504GB",
   "model": "Llama-3.3-70B-Instruct",
   "model_size": null,
   "quant": "Q4_K_M",
   "runtime": "llama.cpp",
   "runtime_version": "0.7.0-dev (build 1, commit ceefd16)",
   "workload": "agent-trace",
   "suite_version": "suite-v1",
   "decode_tps": 12.72,
   "prefill_tps": 1574.48,
   "ttft_ms": 1343.9,
   "prompt_tokens": 3235,
   "output_tokens": 840,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_z-l1yelv_67",
   "run_url": "https://llm-speed.com/r/r_z-l1yelv_67",
   "date": "2026-10-15",
   "hardware": "RTX 5090 (32GB) + AMD EPYC 9534 64-Core Processor (128c) + 756GB",
   "model": "NVIDIA-Nemotron-3.5-Bolt-74B-A9B-24GB",
   "model_size": null,
   "quant": null,
   "runtime": "llama.cpp",
   "runtime_version": "0.7.0-dev (build 11669, commit 620d2225c)",
   "workload": "concurrent-decode",
   "suite_version": "suite-v1",
   "decode_tps": 155.98,
   "prefill_tps": null,
   "ttft_ms": null,
   "prompt_tokens": 1024,
   "output_tokens": 256,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_z-l1yelv_67",
   "run_url": "https://llm-speed.com/r/r_z-l1yelv_67",
   "date": "2026-10-15",
   "hardware": "RTX 5090 (32GB) + AMD EPYC 9534 64-Core Processor (128c) + 756GB",
   "model": "NVIDIA-Nemotron-3.5-Bolt-74B-A9B-24GB",
   "model_size": null,
   "quant": null,
   "runtime": "llama.cpp",
   "runtime_version": "0.7.0-dev (build 11669, commit 620d2225c)",
   "workload": "chat-short",
   "suite_version": "suite-v1",
   "decode_tps": 154.74,
   "prefill_tps": 353.4,
   "ttft_ms": 339.6,
   "prompt_tokens": 120,
   "output_tokens": 256,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_z-l1yelv_67",
   "run_url": "https://llm-speed.com/r/r_z-l1yelv_67",
   "date": "2026-10-15",
   "hardware": "RTX 5090 (32GB) + AMD EPYC 9534 64-Core Processor (128c) + 756GB",
   "model": "NVIDIA-Nemotron-3.5-Bolt-74B-A9B-24GB",
   "model_size": null,
   "quant": null,
   "runtime": "llama.cpp",
   "runtime_version": "0.7.0-dev (build 11669, commit 620d2225c)",
   "workload": "chat-long",
   "suite_version": "suite-v1",
   "decode_tps": 154.23,
   "prefill_tps": 2623.3,
   "ttft_ms": 1238.5,
   "prompt_tokens": 3249,
   "output_tokens": 847,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_z-l1yelv_67",
   "run_url": "https://llm-speed.com/r/r_z-l1yelv_67",
   "date": "2026-10-15",
   "hardware": "RTX 5090 (32GB) + AMD EPYC 9534 64-Core Processor (128c) + 756GB",
   "model": "NVIDIA-Nemotron-3.5-Bolt-74B-A9B-24GB",
   "model_size": null,
   "quant": null,
   "runtime": "llama.cpp",
   "runtime_version": "0.7.0-dev (build 11669, commit 620d2225c)",
   "workload": "agent-trace",
   "suite_version": "suite-v1",
   "decode_tps": 149.32,
   "prefill_tps": 5097.22,
   "ttft_ms": 418.1,
   "prompt_tokens": 3449,
   "output_tokens": 840,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_ymx58emedxd",
   "run_url": "https://llm-speed.com/r/r_ymx58emedxd",
   "date": "2026-10-15",
   "hardware": "RTX 5090 (32GB) + AMD EPYC 9534 64-Core Processor (128c) + 756GB",
   "model": "NVIDIA-Nemotron-3.5-Bolt-74B-A9B-24GB-ctx32k",
   "model_size": null,
   "quant": null,
   "runtime": "llama.cpp",
   "runtime_version": "0.7.0-dev (build 11669, commit 620d2225c)",
   "workload": "concurrent-decode",
   "suite_version": "suite-v1",
   "decode_tps": 157.72,
   "prefill_tps": null,
   "ttft_ms": null,
   "prompt_tokens": 1024,
   "output_tokens": 256,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_ymx58emedxd",
   "run_url": "https://llm-speed.com/r/r_ymx58emedxd",
   "date": "2026-10-15",
   "hardware": "RTX 5090 (32GB) + AMD EPYC 9534 64-Core Processor (128c) + 756GB",
   "model": "NVIDIA-Nemotron-3.5-Bolt-74B-A9B-24GB-ctx32k",
   "model_size": null,
   "quant": null,
   "runtime": "llama.cpp",
   "runtime_version": "0.7.0-dev (build 11669, commit 620d2225c)",
   "workload": "chat-short",
   "suite_version": "suite-v1",
   "decode_tps": 157.26,
   "prefill_tps": 355.42,
   "ttft_ms": 337.6,
   "prompt_tokens": 120,
   "output_tokens": 256,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_ymx58emedxd",
   "run_url": "https://llm-speed.com/r/r_ymx58emedxd",
   "date": "2026-10-15",
   "hardware": "RTX 5090 (32GB) + AMD EPYC 9534 64-Core Processor (128c) + 756GB",
   "model": "NVIDIA-Nemotron-3.5-Bolt-74B-A9B-24GB-ctx32k",
   "model_size": null,
   "quant": null,
   "runtime": "llama.cpp",
   "runtime_version": "0.7.0-dev (build 11669, commit 620d2225c)",
   "workload": "chat-long",
   "suite_version": "suite-v1",
   "decode_tps": 155.74,
   "prefill_tps": 2605.82,
   "ttft_ms": 1246.8,
   "prompt_tokens": 3249,
   "output_tokens": 1024,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_ymx58emedxd",
   "run_url": "https://llm-speed.com/r/r_ymx58emedxd",
   "date": "2026-10-15",
   "hardware": "RTX 5090 (32GB) + AMD EPYC 9534 64-Core Processor (128c) + 756GB",
   "model": "NVIDIA-Nemotron-3.5-Bolt-74B-A9B-24GB-ctx32k",
   "model_size": null,
   "quant": null,
   "runtime": "llama.cpp",
   "runtime_version": "0.7.0-dev (build 11669, commit 620d2225c)",
   "workload": "agent-trace",
   "suite_version": "suite-v1",
   "decode_tps": 154.85,
   "prefill_tps": 5095.71,
   "ttft_ms": 431.3,
   "prompt_tokens": 3449,
   "output_tokens": 840,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_w3lond-s75t",
   "run_url": "https://llm-speed.com/r/r_w3lond-s75t",
   "date": "2026-10-15",
   "hardware": "RTX 3090 (24GB) + AMD EPYC 7C13 64-Core Processor (64c) + 252GB",
   "model": "NVIDIA-Nemotron-3.5-Bolt-74B-A9B-24GB-ctx8k",
   "model_size": null,
   "quant": null,
   "runtime": "llama.cpp",
   "runtime_version": "0.7.0-dev (build 11669, commit 620d2225c)",
   "workload": "concurrent-decode",
   "suite_version": "suite-v1",
   "decode_tps": 80.44,
   "prefill_tps": null,
   "ttft_ms": null,
   "prompt_tokens": 1024,
   "output_tokens": 256,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_w3lond-s75t",
   "run_url": "https://llm-speed.com/r/r_w3lond-s75t",
   "date": "2026-10-15",
   "hardware": "RTX 3090 (24GB) + AMD EPYC 7C13 64-Core Processor (64c) + 252GB",
   "model": "NVIDIA-Nemotron-3.5-Bolt-74B-A9B-24GB-ctx8k",
   "model_size": null,
   "quant": null,
   "runtime": "llama.cpp",
   "runtime_version": "0.7.0-dev (build 11669, commit 620d2225c)",
   "workload": "chat-short",
   "suite_version": "suite-v1",
   "decode_tps": 80.32,
   "prefill_tps": 264.29,
   "ttft_ms": 454.1,
   "prompt_tokens": 120,
   "output_tokens": 256,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_w3lond-s75t",
   "run_url": "https://llm-speed.com/r/r_w3lond-s75t",
   "date": "2026-10-15",
   "hardware": "RTX 3090 (24GB) + AMD EPYC 7C13 64-Core Processor (64c) + 252GB",
   "model": "NVIDIA-Nemotron-3.5-Bolt-74B-A9B-24GB-ctx8k",
   "model_size": null,
   "quant": null,
   "runtime": "llama.cpp",
   "runtime_version": "0.7.0-dev (build 11669, commit 620d2225c)",
   "workload": "chat-long",
   "suite_version": "suite-v1",
   "decode_tps": 80.36,
   "prefill_tps": 1072.23,
   "ttft_ms": 3030.1,
   "prompt_tokens": 3249,
   "output_tokens": 1024,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_w3lond-s75t",
   "run_url": "https://llm-speed.com/r/r_w3lond-s75t",
   "date": "2026-10-15",
   "hardware": "RTX 3090 (24GB) + AMD EPYC 7C13 64-Core Processor (64c) + 252GB",
   "model": "NVIDIA-Nemotron-3.5-Bolt-74B-A9B-24GB-ctx8k",
   "model_size": null,
   "quant": null,
   "runtime": "llama.cpp",
   "runtime_version": "0.7.0-dev (build 11669, commit 620d2225c)",
   "workload": "agent-trace",
   "suite_version": "suite-v1",
   "decode_tps": 79.25,
   "prefill_tps": 2109.32,
   "ttft_ms": 943.9,
   "prompt_tokens": 3449,
   "output_tokens": 840,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_vkjsinuhlyl",
   "run_url": "https://llm-speed.com/r/r_vkjsinuhlyl",
   "date": "2026-10-15",
   "hardware": "RTX 4090 (24GB) + AMD EPYC 7642 48-Core Processor (48c) + 252GB",
   "model": "NVIDIA-Nemotron-3.5-Bolt-74B-A9B-24GB",
   "model_size": null,
   "quant": null,
   "runtime": "llama.cpp",
   "runtime_version": "0.7.0-dev (build 11669, commit 620d2225c)",
   "workload": "concurrent-decode",
   "suite_version": "suite-v1",
   "decode_tps": 106.74,
   "prefill_tps": null,
   "ttft_ms": null,
   "prompt_tokens": 1024,
   "output_tokens": 256,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_vkjsinuhlyl",
   "run_url": "https://llm-speed.com/r/r_vkjsinuhlyl",
   "date": "2026-10-15",
   "hardware": "RTX 4090 (24GB) + AMD EPYC 7642 48-Core Processor (48c) + 252GB",
   "model": "NVIDIA-Nemotron-3.5-Bolt-74B-A9B-24GB",
   "model_size": null,
   "quant": null,
   "runtime": "llama.cpp",
   "runtime_version": "0.7.0-dev (build 11669, commit 620d2225c)",
   "workload": "chat-short",
   "suite_version": "suite-v1",
   "decode_tps": 107.33,
   "prefill_tps": 255.8,
   "ttft_ms": 469.1,
   "prompt_tokens": 120,
   "output_tokens": 256,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_vkjsinuhlyl",
   "run_url": "https://llm-speed.com/r/r_vkjsinuhlyl",
   "date": "2026-10-15",
   "hardware": "RTX 4090 (24GB) + AMD EPYC 7642 48-Core Processor (48c) + 252GB",
   "model": "NVIDIA-Nemotron-3.5-Bolt-74B-A9B-24GB",
   "model_size": null,
   "quant": null,
   "runtime": "llama.cpp",
   "runtime_version": "0.7.0-dev (build 11669, commit 620d2225c)",
   "workload": "chat-long",
   "suite_version": "suite-v1",
   "decode_tps": 106.13,
   "prefill_tps": 1814.72,
   "ttft_ms": 1790.4,
   "prompt_tokens": 3249,
   "output_tokens": 847,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_vkjsinuhlyl",
   "run_url": "https://llm-speed.com/r/r_vkjsinuhlyl",
   "date": "2026-10-15",
   "hardware": "RTX 4090 (24GB) + AMD EPYC 7642 48-Core Processor (48c) + 252GB",
   "model": "NVIDIA-Nemotron-3.5-Bolt-74B-A9B-24GB",
   "model_size": null,
   "quant": null,
   "runtime": "llama.cpp",
   "runtime_version": "0.7.0-dev (build 11669, commit 620d2225c)",
   "workload": "agent-trace",
   "suite_version": "suite-v1",
   "decode_tps": 103.93,
   "prefill_tps": 3318.86,
   "ttft_ms": 636.4,
   "prompt_tokens": 3449,
   "output_tokens": 840,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_qibhq5bkmgq",
   "run_url": "https://llm-speed.com/r/r_qibhq5bkmgq",
   "date": "2026-10-15",
   "hardware": "RTX 4090 (24GB) + AMD EPYC 7642 48-Core Processor (48c) + 252GB",
   "model": "NVIDIA-Nemotron-3.5-Bolt-74B-A9B-24GB-ctx16k",
   "model_size": null,
   "quant": null,
   "runtime": "llama.cpp",
   "runtime_version": "0.7.0-dev (build 11669, commit 620d2225c)",
   "workload": "concurrent-decode",
   "suite_version": "suite-v1",
   "decode_tps": 105.77,
   "prefill_tps": null,
   "ttft_ms": null,
   "prompt_tokens": 1024,
   "output_tokens": 256,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_qibhq5bkmgq",
   "run_url": "https://llm-speed.com/r/r_qibhq5bkmgq",
   "date": "2026-10-15",
   "hardware": "RTX 4090 (24GB) + AMD EPYC 7642 48-Core Processor (48c) + 252GB",
   "model": "NVIDIA-Nemotron-3.5-Bolt-74B-A9B-24GB-ctx16k",
   "model_size": null,
   "quant": null,
   "runtime": "llama.cpp",
   "runtime_version": "0.7.0-dev (build 11669, commit 620d2225c)",
   "workload": "chat-short",
   "suite_version": "suite-v1",
   "decode_tps": 105.7,
   "prefill_tps": 252.84,
   "ttft_ms": 474.6,
   "prompt_tokens": 120,
   "output_tokens": 256,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_qibhq5bkmgq",
   "run_url": "https://llm-speed.com/r/r_qibhq5bkmgq",
   "date": "2026-10-15",
   "hardware": "RTX 4090 (24GB) + AMD EPYC 7642 48-Core Processor (48c) + 252GB",
   "model": "NVIDIA-Nemotron-3.5-Bolt-74B-A9B-24GB-ctx16k",
   "model_size": null,
   "quant": null,
   "runtime": "llama.cpp",
   "runtime_version": "0.7.0-dev (build 11669, commit 620d2225c)",
   "workload": "chat-long",
   "suite_version": "suite-v1",
   "decode_tps": 105.16,
   "prefill_tps": 1746.2,
   "ttft_ms": 1860.6,
   "prompt_tokens": 3249,
   "output_tokens": 1024,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_qibhq5bkmgq",
   "run_url": "https://llm-speed.com/r/r_qibhq5bkmgq",
   "date": "2026-10-15",
   "hardware": "RTX 4090 (24GB) + AMD EPYC 7642 48-Core Processor (48c) + 252GB",
   "model": "NVIDIA-Nemotron-3.5-Bolt-74B-A9B-24GB-ctx16k",
   "model_size": null,
   "quant": null,
   "runtime": "llama.cpp",
   "runtime_version": "0.7.0-dev (build 11669, commit 620d2225c)",
   "workload": "agent-trace",
   "suite_version": "suite-v1",
   "decode_tps": 102.02,
   "prefill_tps": 3280.72,
   "ttft_ms": 673.0,
   "prompt_tokens": 3449,
   "output_tokens": 840,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_qfrb2zmyk0k",
   "run_url": "https://llm-speed.com/r/r_qfrb2zmyk0k",
   "date": "2026-10-15",
   "hardware": "RTX 5090 (32GB) + AMD EPYC 9534 64-Core Processor (128c) + 756GB",
   "model": "Qwen3.8-27B",
   "model_size": null,
   "quant": "Q4_K_M",
   "runtime": "llama.cpp",
   "runtime_version": "0.7.0-dev (build 11669, commit 620d2225c)",
   "workload": "concurrent-decode",
   "suite_version": "suite-v1",
   "decode_tps": 75.4,
   "prefill_tps": null,
   "ttft_ms": null,
   "prompt_tokens": 1024,
   "output_tokens": 256,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_qfrb2zmyk0k",
   "run_url": "https://llm-speed.com/r/r_qfrb2zmyk0k",
   "date": "2026-10-15",
   "hardware": "RTX 5090 (32GB) + AMD EPYC 9534 64-Core Processor (128c) + 756GB",
   "model": "Qwen3.8-27B",
   "model_size": null,
   "quant": "Q4_K_M",
   "runtime": "llama.cpp",
   "runtime_version": "0.7.0-dev (build 11669, commit 620d2225c)",
   "workload": "chat-short",
   "suite_version": "suite-v1",
   "decode_tps": 75.19,
   "prefill_tps": 563.37,
   "ttft_ms": 273.4,
   "prompt_tokens": 154,
   "output_tokens": 256,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_qfrb2zmyk0k",
   "run_url": "https://llm-speed.com/r/r_qfrb2zmyk0k",
   "date": "2026-10-15",
   "hardware": "RTX 5090 (32GB) + AMD EPYC 9534 64-Core Processor (128c) + 756GB",
   "model": "Qwen3.8-27B",
   "model_size": null,
   "quant": "Q4_K_M",
   "runtime": "llama.cpp",
   "runtime_version": "0.7.0-dev (build 11669, commit 620d2225c)",
   "workload": "chat-long",
   "suite_version": "suite-v1",
   "decode_tps": 74.57,
   "prefill_tps": 2499.5,
   "ttft_ms": 1289.9,
   "prompt_tokens": 3224,
   "output_tokens": 872,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_qfrb2zmyk0k",
   "run_url": "https://llm-speed.com/r/r_qfrb2zmyk0k",
   "date": "2026-10-15",
   "hardware": "RTX 5090 (32GB) + AMD EPYC 9534 64-Core Processor (128c) + 756GB",
   "model": "Qwen3.8-27B",
   "model_size": null,
   "quant": "Q4_K_M",
   "runtime": "llama.cpp",
   "runtime_version": "0.7.0-dev (build 11669, commit 620d2225c)",
   "workload": "agent-trace",
   "suite_version": "suite-v1",
   "decode_tps": 73.52,
   "prefill_tps": 5065.29,
   "ttft_ms": 436.1,
   "prompt_tokens": 3496,
   "output_tokens": 840,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_j5tgvia-kjk",
   "run_url": "https://llm-speed.com/r/r_j5tgvia-kjk",
   "date": "2026-10-15",
   "hardware": "RTX A4000 (16GB) + AMD EPYC 7402P 24-Core Processor (24c) + 126GB",
   "model": "LFM2.5-24B-A2B",
   "model_size": null,
   "quant": "Q4_K_M",
   "runtime": "llama.cpp",
   "runtime_version": "0.7.0-dev (build 1, commit b76b133)",
   "workload": "concurrent-decode",
   "suite_version": "suite-v1",
   "decode_tps": 142.7,
   "prefill_tps": null,
   "ttft_ms": null,
   "prompt_tokens": 1024,
   "output_tokens": 256,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_j5tgvia-kjk",
   "run_url": "https://llm-speed.com/r/r_j5tgvia-kjk",
   "date": "2026-10-15",
   "hardware": "RTX A4000 (16GB) + AMD EPYC 7402P 24-Core Processor (24c) + 126GB",
   "model": "LFM2.5-24B-A2B",
   "model_size": null,
   "quant": "Q4_K_M",
   "runtime": "llama.cpp",
   "runtime_version": "0.7.0-dev (build 1, commit b76b133)",
   "workload": "chat-short",
   "suite_version": "suite-v1",
   "decode_tps": 140.02,
   "prefill_tps": 453.55,
   "ttft_ms": 251.3,
   "prompt_tokens": 114,
   "output_tokens": 256,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_j5tgvia-kjk",
   "run_url": "https://llm-speed.com/r/r_j5tgvia-kjk",
   "date": "2026-10-15",
   "hardware": "RTX A4000 (16GB) + AMD EPYC 7402P 24-Core Processor (24c) + 126GB",
   "model": "LFM2.5-24B-A2B",
   "model_size": null,
   "quant": "Q4_K_M",
   "runtime": "llama.cpp",
   "runtime_version": "0.7.0-dev (build 1, commit b76b133)",
   "workload": "chat-long",
   "suite_version": "suite-v1",
   "decode_tps": 139.93,
   "prefill_tps": 2684.38,
   "ttft_ms": 1223.4,
   "prompt_tokens": 3284,
   "output_tokens": 1024,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_j5tgvia-kjk",
   "run_url": "https://llm-speed.com/r/r_j5tgvia-kjk",
   "date": "2026-10-15",
   "hardware": "RTX A4000 (16GB) + AMD EPYC 7402P 24-Core Processor (24c) + 126GB",
   "model": "LFM2.5-24B-A2B",
   "model_size": null,
   "quant": "Q4_K_M",
   "runtime": "llama.cpp",
   "runtime_version": "0.7.0-dev (build 1, commit b76b133)",
   "workload": "agent-trace",
   "suite_version": "suite-v1",
   "decode_tps": 139.44,
   "prefill_tps": 3923.48,
   "ttft_ms": 391.1,
   "prompt_tokens": 3514,
   "output_tokens": 840,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_iu0s18cs9d8",
   "run_url": "https://llm-speed.com/r/r_iu0s18cs9d8",
   "date": "2026-10-15",
   "hardware": "RTX 4090 (24GB) + AMD EPYC 7642 48-Core Processor (48c) + 252GB",
   "model": "NVIDIA-Nemotron-3.5-Bolt-74B-A9B-24GB-ctx8k",
   "model_size": null,
   "quant": null,
   "runtime": "llama.cpp",
   "runtime_version": "0.7.0-dev (build 11669, commit 620d2225c)",
   "workload": "concurrent-decode",
   "suite_version": "suite-v1",
   "decode_tps": 106.36,
   "prefill_tps": null,
   "ttft_ms": null,
   "prompt_tokens": 1024,
   "output_tokens": 256,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_iu0s18cs9d8",
   "run_url": "https://llm-speed.com/r/r_iu0s18cs9d8",
   "date": "2026-10-15",
   "hardware": "RTX 4090 (24GB) + AMD EPYC 7642 48-Core Processor (48c) + 252GB",
   "model": "NVIDIA-Nemotron-3.5-Bolt-74B-A9B-24GB-ctx8k",
   "model_size": null,
   "quant": null,
   "runtime": "llama.cpp",
   "runtime_version": "0.7.0-dev (build 11669, commit 620d2225c)",
   "workload": "chat-short",
   "suite_version": "suite-v1",
   "decode_tps": 107.01,
   "prefill_tps": 253.34,
   "ttft_ms": 473.7,
   "prompt_tokens": 120,
   "output_tokens": 256,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_iu0s18cs9d8",
   "run_url": "https://llm-speed.com/r/r_iu0s18cs9d8",
   "date": "2026-10-15",
   "hardware": "RTX 4090 (24GB) + AMD EPYC 7642 48-Core Processor (48c) + 252GB",
   "model": "NVIDIA-Nemotron-3.5-Bolt-74B-A9B-24GB-ctx8k",
   "model_size": null,
   "quant": null,
   "runtime": "llama.cpp",
   "runtime_version": "0.7.0-dev (build 11669, commit 620d2225c)",
   "workload": "chat-long",
   "suite_version": "suite-v1",
   "decode_tps": 105.49,
   "prefill_tps": 1743.8,
   "ttft_ms": 1863.2,
   "prompt_tokens": 3249,
   "output_tokens": 1024,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_iu0s18cs9d8",
   "run_url": "https://llm-speed.com/r/r_iu0s18cs9d8",
   "date": "2026-10-15",
   "hardware": "RTX 4090 (24GB) + AMD EPYC 7642 48-Core Processor (48c) + 252GB",
   "model": "NVIDIA-Nemotron-3.5-Bolt-74B-A9B-24GB-ctx8k",
   "model_size": null,
   "quant": null,
   "runtime": "llama.cpp",
   "runtime_version": "0.7.0-dev (build 11669, commit 620d2225c)",
   "workload": "agent-trace",
   "suite_version": "suite-v1",
   "decode_tps": 103.04,
   "prefill_tps": 3340.41,
   "ttft_ms": 661.0,
   "prompt_tokens": 3449,
   "output_tokens": 840,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_c8k6fleym09",
   "run_url": "https://llm-speed.com/r/r_c8k6fleym09",
   "date": "2026-10-15",
   "hardware": "RTX 5090 (32GB) + INTEL(R) XEON(R) GOLD 6530 (64c) + 1008GB",
   "model": "LFM2.5-24B-A2B",
   "model_size": null,
   "quant": "Q4_K_M",
   "runtime": "llama.cpp",
   "runtime_version": "0.6.0-dev (build 1, commit a0e6dc7)",
   "workload": "concurrent-decode",
   "suite_version": "suite-v1",
   "decode_tps": 321.27,
   "prefill_tps": null,
   "ttft_ms": null,
   "prompt_tokens": 1024,
   "output_tokens": 256,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_c8k6fleym09",
   "run_url": "https://llm-speed.com/r/r_c8k6fleym09",
   "date": "2026-10-15",
   "hardware": "RTX 5090 (32GB) + INTEL(R) XEON(R) GOLD 6530 (64c) + 1008GB",
   "model": "LFM2.5-24B-A2B",
   "model_size": null,
   "quant": "Q4_K_M",
   "runtime": "llama.cpp",
   "runtime_version": "0.6.0-dev (build 1, commit a0e6dc7)",
   "workload": "chat-short",
   "suite_version": "suite-v1",
   "decode_tps": 308.53,
   "prefill_tps": 553.12,
   "ttft_ms": 206.1,
   "prompt_tokens": 114,
   "output_tokens": 256,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_c8k6fleym09",
   "run_url": "https://llm-speed.com/r/r_c8k6fleym09",
   "date": "2026-10-15",
   "hardware": "RTX 5090 (32GB) + INTEL(R) XEON(R) GOLD 6530 (64c) + 1008GB",
   "model": "LFM2.5-24B-A2B",
   "model_size": null,
   "quant": "Q4_K_M",
   "runtime": "llama.cpp",
   "runtime_version": "0.6.0-dev (build 1, commit a0e6dc7)",
   "workload": "chat-long",
   "suite_version": "suite-v1",
   "decode_tps": 300.43,
   "prefill_tps": 12418.32,
   "ttft_ms": 264.4,
   "prompt_tokens": 3284,
   "output_tokens": 1024,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_c8k6fleym09",
   "run_url": "https://llm-speed.com/r/r_c8k6fleym09",
   "date": "2026-10-15",
   "hardware": "RTX 5090 (32GB) + INTEL(R) XEON(R) GOLD 6530 (64c) + 1008GB",
   "model": "LFM2.5-24B-A2B",
   "model_size": null,
   "quant": "Q4_K_M",
   "runtime": "llama.cpp",
   "runtime_version": "0.6.0-dev (build 1, commit a0e6dc7)",
   "workload": "agent-trace",
   "suite_version": "suite-v1",
   "decode_tps": 313.18,
   "prefill_tps": 18108.76,
   "ttft_ms": 116.5,
   "prompt_tokens": 3514,
   "output_tokens": 840,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_bqmpv9kncyd",
   "run_url": "https://llm-speed.com/r/r_bqmpv9kncyd",
   "date": "2026-10-15",
   "hardware": "RTX 4090 (24GB) + AMD EPYC 7452 32-Core Processor (64c) + 1008GB",
   "model": "LFM2.5-24B-A2B",
   "model_size": null,
   "quant": "Q4_K_M",
   "runtime": "llama.cpp",
   "runtime_version": "0.6.0-dev (build 1, commit a0e6dc7)",
   "workload": "concurrent-decode",
   "suite_version": "suite-v1",
   "decode_tps": 236.81,
   "prefill_tps": null,
   "ttft_ms": null,
   "prompt_tokens": 1024,
   "output_tokens": 256,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_bqmpv9kncyd",
   "run_url": "https://llm-speed.com/r/r_bqmpv9kncyd",
   "date": "2026-10-15",
   "hardware": "RTX 4090 (24GB) + AMD EPYC 7452 32-Core Processor (64c) + 1008GB",
   "model": "LFM2.5-24B-A2B",
   "model_size": null,
   "quant": "Q4_K_M",
   "runtime": "llama.cpp",
   "runtime_version": "0.6.0-dev (build 1, commit a0e6dc7)",
   "workload": "chat-short",
   "suite_version": "suite-v1",
   "decode_tps": 231.89,
   "prefill_tps": 640.09,
   "ttft_ms": 178.1,
   "prompt_tokens": 114,
   "output_tokens": 256,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_bqmpv9kncyd",
   "run_url": "https://llm-speed.com/r/r_bqmpv9kncyd",
   "date": "2026-10-15",
   "hardware": "RTX 4090 (24GB) + AMD EPYC 7452 32-Core Processor (64c) + 1008GB",
   "model": "LFM2.5-24B-A2B",
   "model_size": null,
   "quant": "Q4_K_M",
   "runtime": "llama.cpp",
   "runtime_version": "0.6.0-dev (build 1, commit a0e6dc7)",
   "workload": "chat-long",
   "suite_version": "suite-v1",
   "decode_tps": 233.44,
   "prefill_tps": 9512.43,
   "ttft_ms": 345.2,
   "prompt_tokens": 3284,
   "output_tokens": 1024,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_bqmpv9kncyd",
   "run_url": "https://llm-speed.com/r/r_bqmpv9kncyd",
   "date": "2026-10-15",
   "hardware": "RTX 4090 (24GB) + AMD EPYC 7452 32-Core Processor (64c) + 1008GB",
   "model": "LFM2.5-24B-A2B",
   "model_size": null,
   "quant": "Q4_K_M",
   "runtime": "llama.cpp",
   "runtime_version": "0.6.0-dev (build 1, commit a0e6dc7)",
   "workload": "agent-trace",
   "suite_version": "suite-v1",
   "decode_tps": 230.96,
   "prefill_tps": 13801.65,
   "ttft_ms": 153.5,
   "prompt_tokens": 3514,
   "output_tokens": 840,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_-sifd5otbpy",
   "run_url": "https://llm-speed.com/r/r_-sifd5otbpy",
   "date": "2026-10-15",
   "hardware": "RTX 3090 (24GB) + AMD EPYC 7C13 64-Core Processor (64c) + 252GB",
   "model": "NVIDIA-Nemotron-3.5-Bolt-74B-A9B-24GB",
   "model_size": null,
   "quant": null,
   "runtime": "llama.cpp",
   "runtime_version": "0.7.0-dev (build 11669, commit 620d2225c)",
   "workload": "concurrent-decode",
   "suite_version": "suite-v1",
   "decode_tps": 87.76,
   "prefill_tps": null,
   "ttft_ms": null,
   "prompt_tokens": 1024,
   "output_tokens": 256,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_-sifd5otbpy",
   "run_url": "https://llm-speed.com/r/r_-sifd5otbpy",
   "date": "2026-10-15",
   "hardware": "RTX 3090 (24GB) + AMD EPYC 7C13 64-Core Processor (64c) + 252GB",
   "model": "NVIDIA-Nemotron-3.5-Bolt-74B-A9B-24GB",
   "model_size": null,
   "quant": null,
   "runtime": "llama.cpp",
   "runtime_version": "0.7.0-dev (build 11669, commit 620d2225c)",
   "workload": "chat-short",
   "suite_version": "suite-v1",
   "decode_tps": 87.75,
   "prefill_tps": 259.74,
   "ttft_ms": 462.0,
   "prompt_tokens": 120,
   "output_tokens": 256,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_-sifd5otbpy",
   "run_url": "https://llm-speed.com/r/r_-sifd5otbpy",
   "date": "2026-10-15",
   "hardware": "RTX 3090 (24GB) + AMD EPYC 7C13 64-Core Processor (64c) + 252GB",
   "model": "NVIDIA-Nemotron-3.5-Bolt-74B-A9B-24GB",
   "model_size": null,
   "quant": null,
   "runtime": "llama.cpp",
   "runtime_version": "0.7.0-dev (build 11669, commit 620d2225c)",
   "workload": "chat-long",
   "suite_version": "suite-v1",
   "decode_tps": 87.48,
   "prefill_tps": 1183.37,
   "ttft_ms": 2745.5,
   "prompt_tokens": 3249,
   "output_tokens": 847,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_-sifd5otbpy",
   "run_url": "https://llm-speed.com/r/r_-sifd5otbpy",
   "date": "2026-10-15",
   "hardware": "RTX 3090 (24GB) + AMD EPYC 7C13 64-Core Processor (64c) + 252GB",
   "model": "NVIDIA-Nemotron-3.5-Bolt-74B-A9B-24GB",
   "model_size": null,
   "quant": null,
   "runtime": "llama.cpp",
   "runtime_version": "0.7.0-dev (build 11669, commit 620d2225c)",
   "workload": "agent-trace",
   "suite_version": "suite-v1",
   "decode_tps": 78.92,
   "prefill_tps": 2094.96,
   "ttft_ms": 894.9,
   "prompt_tokens": 3449,
   "output_tokens": 840,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_ttniv2j74vk",
   "run_url": "https://llm-speed.com/r/r_ttniv2j74vk",
   "date": "2026-10-14",
   "hardware": "RTX 4090 (24GB) + AMD EPYC 7642 48-Core Processor (48c) + 252GB",
   "model": "qwen2.5-coder",
   "model_size": "14.8B",
   "quant": "Q4_K_M",
   "runtime": "ollama",
   "runtime_version": "0.31.1",
   "workload": "concurrent-decode",
   "suite_version": "suite-v1",
   "decode_tps": 80.11,
   "prefill_tps": null,
   "ttft_ms": null,
   "prompt_tokens": 1024,
   "output_tokens": 256,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_ttniv2j74vk",
   "run_url": "https://llm-speed.com/r/r_ttniv2j74vk",
   "date": "2026-10-14",
   "hardware": "RTX 4090 (24GB) + AMD EPYC 7642 48-Core Processor (48c) + 252GB",
   "model": "qwen2.5-coder",
   "model_size": "14.8B",
   "quant": "Q4_K_M",
   "runtime": "ollama",
   "runtime_version": "0.31.1",
   "workload": "chat-short",
   "suite_version": "suite-v1",
   "decode_tps": 81.18,
   "prefill_tps": 324.39,
   "ttft_ms": 403.8,
   "prompt_tokens": 131,
   "output_tokens": 256,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_ttniv2j74vk",
   "run_url": "https://llm-speed.com/r/r_ttniv2j74vk",
   "date": "2026-10-14",
   "hardware": "RTX 4090 (24GB) + AMD EPYC 7642 48-Core Processor (48c) + 252GB",
   "model": "qwen2.5-coder",
   "model_size": "14.8B",
   "quant": "Q4_K_M",
   "runtime": "ollama",
   "runtime_version": "0.31.1",
   "workload": "chat-long",
   "suite_version": "suite-v1",
   "decode_tps": 76.65,
   "prefill_tps": 3252.19,
   "ttft_ms": 974.1,
   "prompt_tokens": 3168,
   "output_tokens": 603,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_ttniv2j74vk",
   "run_url": "https://llm-speed.com/r/r_ttniv2j74vk",
   "date": "2026-10-14",
   "hardware": "RTX 4090 (24GB) + AMD EPYC 7642 48-Core Processor (48c) + 252GB",
   "model": "qwen2.5-coder",
   "model_size": "14.8B",
   "quant": "Q4_K_M",
   "runtime": "ollama",
   "runtime_version": "0.31.1",
   "workload": "agent-trace",
   "suite_version": "suite-v1",
   "decode_tps": 78.63,
   "prefill_tps": 4874.59,
   "ttft_ms": 448.6,
   "prompt_tokens": 3268,
   "output_tokens": 840,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_t-a9v_ipdit",
   "run_url": "https://llm-speed.com/r/r_t-a9v_ipdit",
   "date": "2026-10-14",
   "hardware": "RTX 4090 (24GB) + AMD EPYC 7642 48-Core Processor (48c) + 252GB",
   "model": "gpt-oss-20b",
   "model_size": null,
   "quant": "Q4_K_M",
   "runtime": "llama.cpp",
   "runtime_version": "0.6.0-dev (build 1, commit df18bee)",
   "workload": "concurrent-decode",
   "suite_version": "suite-v1",
   "decode_tps": 239.65,
   "prefill_tps": null,
   "ttft_ms": null,
   "prompt_tokens": 1024,
   "output_tokens": 256,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_t-a9v_ipdit",
   "run_url": "https://llm-speed.com/r/r_t-a9v_ipdit",
   "date": "2026-10-14",
   "hardware": "RTX 4090 (24GB) + AMD EPYC 7642 48-Core Processor (48c) + 252GB",
   "model": "gpt-oss-20b",
   "model_size": null,
   "quant": "Q4_K_M",
   "runtime": "llama.cpp",
   "runtime_version": "0.6.0-dev (build 1, commit df18bee)",
   "workload": "chat-short",
   "suite_version": "suite-v1",
   "decode_tps": 242.65,
   "prefill_tps": 593.85,
   "ttft_ms": 279.5,
   "prompt_tokens": 166,
   "output_tokens": 256,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_t-a9v_ipdit",
   "run_url": "https://llm-speed.com/r/r_t-a9v_ipdit",
   "date": "2026-10-14",
   "hardware": "RTX 4090 (24GB) + AMD EPYC 7642 48-Core Processor (48c) + 252GB",
   "model": "gpt-oss-20b",
   "model_size": null,
   "quant": "Q4_K_M",
   "runtime": "llama.cpp",
   "runtime_version": "0.6.0-dev (build 1, commit df18bee)",
   "workload": "chat-long",
   "suite_version": "suite-v1",
   "decode_tps": 231.21,
   "prefill_tps": 7864.1,
   "ttft_ms": 405.4,
   "prompt_tokens": 3188,
   "output_tokens": 1024,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_t-a9v_ipdit",
   "run_url": "https://llm-speed.com/r/r_t-a9v_ipdit",
   "date": "2026-10-14",
   "hardware": "RTX 4090 (24GB) + AMD EPYC 7642 48-Core Processor (48c) + 252GB",
   "model": "gpt-oss-20b",
   "model_size": null,
   "quant": "Q4_K_M",
   "runtime": "llama.cpp",
   "runtime_version": "0.6.0-dev (build 1, commit df18bee)",
   "workload": "agent-trace",
   "suite_version": "suite-v1",
   "decode_tps": 236.11,
   "prefill_tps": 13439.22,
   "ttft_ms": 150.3,
   "prompt_tokens": 3287,
   "output_tokens": 840,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_ptct8o01o5l",
   "run_url": "https://llm-speed.com/r/r_ptct8o01o5l",
   "date": "2026-10-14",
   "hardware": "RTX 3090 (24GB) + AMD EPYC 7C13 64-Core Processor (128c) + 1008GB",
   "model": "Qwen3.8-27B",
   "model_size": null,
   "quant": "Q4_K_M",
   "runtime": "llama.cpp",
   "runtime_version": "0.6.0-dev (build 1, commit 8c2134a)",
   "workload": "concurrent-decode",
   "suite_version": "suite-v1",
   "decode_tps": 37.53,
   "prefill_tps": null,
   "ttft_ms": null,
   "prompt_tokens": 1024,
   "output_tokens": 256,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_ptct8o01o5l",
   "run_url": "https://llm-speed.com/r/r_ptct8o01o5l",
   "date": "2026-10-14",
   "hardware": "RTX 3090 (24GB) + AMD EPYC 7C13 64-Core Processor (128c) + 1008GB",
   "model": "Qwen3.8-27B",
   "model_size": null,
   "quant": "Q4_K_M",
   "runtime": "llama.cpp",
   "runtime_version": "0.6.0-dev (build 1, commit 8c2134a)",
   "workload": "chat-short",
   "suite_version": "suite-v1",
   "decode_tps": 37.65,
   "prefill_tps": 368.52,
   "ttft_ms": 417.9,
   "prompt_tokens": 154,
   "output_tokens": 256,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_ptct8o01o5l",
   "run_url": "https://llm-speed.com/r/r_ptct8o01o5l",
   "date": "2026-10-14",
   "hardware": "RTX 3090 (24GB) + AMD EPYC 7C13 64-Core Processor (128c) + 1008GB",
   "model": "Qwen3.8-27B",
   "model_size": null,
   "quant": "Q4_K_M",
   "runtime": "llama.cpp",
   "runtime_version": "0.6.0-dev (build 1, commit 8c2134a)",
   "workload": "chat-long",
   "suite_version": "suite-v1",
   "decode_tps": 37.25,
   "prefill_tps": 1103.55,
   "ttft_ms": 2921.5,
   "prompt_tokens": 3224,
   "output_tokens": 1024,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_ptct8o01o5l",
   "run_url": "https://llm-speed.com/r/r_ptct8o01o5l",
   "date": "2026-10-14",
   "hardware": "RTX 3090 (24GB) + AMD EPYC 7C13 64-Core Processor (128c) + 1008GB",
   "model": "Qwen3.8-27B",
   "model_size": null,
   "quant": "Q4_K_M",
   "runtime": "llama.cpp",
   "runtime_version": "0.6.0-dev (build 1, commit 8c2134a)",
   "workload": "agent-trace",
   "suite_version": "suite-v1",
   "decode_tps": 36.75,
   "prefill_tps": 2190.51,
   "ttft_ms": 967.9,
   "prompt_tokens": 3496,
   "output_tokens": 840,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_nrpc8-onxu2",
   "run_url": "https://llm-speed.com/r/r_nrpc8-onxu2",
   "date": "2026-10-14",
   "hardware": "RTX 4090 (24GB) + AMD EPYC 7642 48-Core Processor (48c) + 252GB",
   "model": "qwen2.5-coder",
   "model_size": "7.6B",
   "quant": "Q4_K_M",
   "runtime": "ollama",
   "runtime_version": "0.31.1",
   "workload": "concurrent-decode",
   "suite_version": "suite-v1",
   "decode_tps": 137.26,
   "prefill_tps": null,
   "ttft_ms": null,
   "prompt_tokens": 1024,
   "output_tokens": 256,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_nrpc8-onxu2",
   "run_url": "https://llm-speed.com/r/r_nrpc8-onxu2",
   "date": "2026-10-14",
   "hardware": "RTX 4090 (24GB) + AMD EPYC 7642 48-Core Processor (48c) + 252GB",
   "model": "qwen2.5-coder",
   "model_size": "7.6B",
   "quant": "Q4_K_M",
   "runtime": "ollama",
   "runtime_version": "0.31.1",
   "workload": "chat-short",
   "suite_version": "suite-v1",
   "decode_tps": 131.56,
   "prefill_tps": 446.19,
   "ttft_ms": 293.6,
   "prompt_tokens": 131,
   "output_tokens": 256,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_nrpc8-onxu2",
   "run_url": "https://llm-speed.com/r/r_nrpc8-onxu2",
   "date": "2026-10-14",
   "hardware": "RTX 4090 (24GB) + AMD EPYC 7642 48-Core Processor (48c) + 252GB",
   "model": "qwen2.5-coder",
   "model_size": "7.6B",
   "quant": "Q4_K_M",
   "runtime": "ollama",
   "runtime_version": "0.31.1",
   "workload": "chat-long",
   "suite_version": "suite-v1",
   "decode_tps": 133.15,
   "prefill_tps": 5116.71,
   "ttft_ms": 619.1,
   "prompt_tokens": 3168,
   "output_tokens": 467,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_nrpc8-onxu2",
   "run_url": "https://llm-speed.com/r/r_nrpc8-onxu2",
   "date": "2026-10-14",
   "hardware": "RTX 4090 (24GB) + AMD EPYC 7642 48-Core Processor (48c) + 252GB",
   "model": "qwen2.5-coder",
   "model_size": "7.6B",
   "quant": "Q4_K_M",
   "runtime": "ollama",
   "runtime_version": "0.31.1",
   "workload": "agent-trace",
   "suite_version": "suite-v1",
   "decode_tps": 135.75,
   "prefill_tps": 5297.29,
   "ttft_ms": 351.8,
   "prompt_tokens": 3268,
   "output_tokens": 840,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_eqnsqmgz29a",
   "run_url": "https://llm-speed.com/r/r_eqnsqmgz29a",
   "date": "2026-10-14",
   "hardware": "RTX 4090 (24GB) + AMD EPYC 7642 48-Core Processor (48c) + 252GB",
   "model": "Qwen2.5-Coder-14B-Instruct",
   "model_size": null,
   "quant": "Q4_K_M",
   "runtime": "llama.cpp",
   "runtime_version": "0.6.0-dev (build 1, commit df18bee)",
   "workload": "concurrent-decode",
   "suite_version": "suite-v1",
   "decode_tps": 88.9,
   "prefill_tps": null,
   "ttft_ms": null,
   "prompt_tokens": 1024,
   "output_tokens": 256,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_eqnsqmgz29a",
   "run_url": "https://llm-speed.com/r/r_eqnsqmgz29a",
   "date": "2026-10-14",
   "hardware": "RTX 4090 (24GB) + AMD EPYC 7642 48-Core Processor (48c) + 252GB",
   "model": "Qwen2.5-Coder-14B-Instruct",
   "model_size": null,
   "quant": "Q4_K_M",
   "runtime": "llama.cpp",
   "runtime_version": "0.6.0-dev (build 1, commit df18bee)",
   "workload": "chat-short",
   "suite_version": "suite-v1",
   "decode_tps": 89.87,
   "prefill_tps": 2292.58,
   "ttft_ms": 57.1,
   "prompt_tokens": 131,
   "output_tokens": 256,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_eqnsqmgz29a",
   "run_url": "https://llm-speed.com/r/r_eqnsqmgz29a",
   "date": "2026-10-14",
   "hardware": "RTX 4090 (24GB) + AMD EPYC 7642 48-Core Processor (48c) + 252GB",
   "model": "Qwen2.5-Coder-14B-Instruct",
   "model_size": null,
   "quant": "Q4_K_M",
   "runtime": "llama.cpp",
   "runtime_version": "0.6.0-dev (build 1, commit df18bee)",
   "workload": "chat-long",
   "suite_version": "suite-v1",
   "decode_tps": 84.17,
   "prefill_tps": 4989.09,
   "ttft_ms": 635.0,
   "prompt_tokens": 3168,
   "output_tokens": 426,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_eqnsqmgz29a",
   "run_url": "https://llm-speed.com/r/r_eqnsqmgz29a",
   "date": "2026-10-14",
   "hardware": "RTX 4090 (24GB) + AMD EPYC 7642 48-Core Processor (48c) + 252GB",
   "model": "Qwen2.5-Coder-14B-Instruct",
   "model_size": null,
   "quant": "Q4_K_M",
   "runtime": "llama.cpp",
   "runtime_version": "0.6.0-dev (build 1, commit df18bee)",
   "workload": "agent-trace",
   "suite_version": "suite-v1",
   "decode_tps": 86.64,
   "prefill_tps": 14684.01,
   "ttft_ms": 142.4,
   "prompt_tokens": 3268,
   "output_tokens": 840,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_d_wda_rw62f",
   "run_url": "https://llm-speed.com/r/r_d_wda_rw62f",
   "date": "2026-10-14",
   "hardware": "RTX 4090 (24GB) + AMD EPYC 7763 64-Core Processor (128c) + 1008GB",
   "model": "Qwen3.6-27B",
   "model_size": null,
   "quant": "Q4_K_M",
   "runtime": "llama.cpp",
   "runtime_version": "0.6.0-dev (build 1, commit 3fb4767)",
   "workload": "concurrent-decode",
   "suite_version": "suite-v1",
   "decode_tps": 47.15,
   "prefill_tps": null,
   "ttft_ms": null,
   "prompt_tokens": 1024,
   "output_tokens": 256,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_d_wda_rw62f",
   "run_url": "https://llm-speed.com/r/r_d_wda_rw62f",
   "date": "2026-10-14",
   "hardware": "RTX 4090 (24GB) + AMD EPYC 7763 64-Core Processor (128c) + 1008GB",
   "model": "Qwen3.6-27B",
   "model_size": null,
   "quant": "Q4_K_M",
   "runtime": "llama.cpp",
   "runtime_version": "0.6.0-dev (build 1, commit 3fb4767)",
   "workload": "chat-short",
   "suite_version": "suite-v1",
   "decode_tps": 47.14,
   "prefill_tps": 349.48,
   "ttft_ms": 320.5,
   "prompt_tokens": 112,
   "output_tokens": 256,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_d_wda_rw62f",
   "run_url": "https://llm-speed.com/r/r_d_wda_rw62f",
   "date": "2026-10-14",
   "hardware": "RTX 4090 (24GB) + AMD EPYC 7763 64-Core Processor (128c) + 1008GB",
   "model": "Qwen3.6-27B",
   "model_size": null,
   "quant": "Q4_K_M",
   "runtime": "llama.cpp",
   "runtime_version": "0.6.0-dev (build 1, commit 3fb4767)",
   "workload": "chat-long",
   "suite_version": "suite-v1",
   "decode_tps": 46.57,
   "prefill_tps": 2107.9,
   "ttft_ms": 1509.6,
   "prompt_tokens": 3182,
   "output_tokens": 1024,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_d_wda_rw62f",
   "run_url": "https://llm-speed.com/r/r_d_wda_rw62f",
   "date": "2026-10-14",
   "hardware": "RTX 4090 (24GB) + AMD EPYC 7763 64-Core Processor (128c) + 1008GB",
   "model": "Qwen3.6-27B",
   "model_size": null,
   "quant": "Q4_K_M",
   "runtime": "llama.cpp",
   "runtime_version": "0.6.0-dev (build 1, commit 3fb4767)",
   "workload": "agent-trace",
   "suite_version": "suite-v1",
   "decode_tps": 46.19,
   "prefill_tps": 3445.65,
   "ttft_ms": 626.2,
   "prompt_tokens": 3454,
   "output_tokens": 840,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_ayy9rjgha5r",
   "run_url": "https://llm-speed.com/r/r_ayy9rjgha5r",
   "date": "2026-10-14",
   "hardware": "RTX 4090 (24GB) + AMD EPYC 7642 48-Core Processor (48c) + 252GB",
   "model": "Qwen3-32B",
   "model_size": null,
   "quant": "Q4_K_M",
   "runtime": "llama.cpp",
   "runtime_version": "0.6.0-dev (build 1, commit df18bee)",
   "workload": "concurrent-decode",
   "suite_version": "suite-v1",
   "decode_tps": 42.69,
   "prefill_tps": null,
   "ttft_ms": null,
   "prompt_tokens": 1024,
   "output_tokens": 256,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_ayy9rjgha5r",
   "run_url": "https://llm-speed.com/r/r_ayy9rjgha5r",
   "date": "2026-10-14",
   "hardware": "RTX 4090 (24GB) + AMD EPYC 7642 48-Core Processor (48c) + 252GB",
   "model": "Qwen3-32B",
   "model_size": null,
   "quant": "Q4_K_M",
   "runtime": "llama.cpp",
   "runtime_version": "0.6.0-dev (build 1, commit df18bee)",
   "workload": "chat-short",
   "suite_version": "suite-v1",
   "decode_tps": 43.47,
   "prefill_tps": 778.56,
   "ttft_ms": 141.3,
   "prompt_tokens": 110,
   "output_tokens": 256,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_ayy9rjgha5r",
   "run_url": "https://llm-speed.com/r/r_ayy9rjgha5r",
   "date": "2026-10-14",
   "hardware": "RTX 4090 (24GB) + AMD EPYC 7642 48-Core Processor (48c) + 252GB",
   "model": "Qwen3-32B",
   "model_size": null,
   "quant": "Q4_K_M",
   "runtime": "llama.cpp",
   "runtime_version": "0.6.0-dev (build 1, commit df18bee)",
   "workload": "chat-long",
   "suite_version": "suite-v1",
   "decode_tps": 40.66,
   "prefill_tps": 2333.26,
   "ttft_ms": 1348.8,
   "prompt_tokens": 3147,
   "output_tokens": 1024,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_ayy9rjgha5r",
   "run_url": "https://llm-speed.com/r/r_ayy9rjgha5r",
   "date": "2026-10-14",
   "hardware": "RTX 4090 (24GB) + AMD EPYC 7642 48-Core Processor (48c) + 252GB",
   "model": "Qwen3-32B",
   "model_size": null,
   "quant": "Q4_K_M",
   "runtime": "llama.cpp",
   "runtime_version": "0.6.0-dev (build 1, commit df18bee)",
   "workload": "agent-trace",
   "suite_version": "suite-v1",
   "decode_tps": 41.94,
   "prefill_tps": 5982.29,
   "ttft_ms": 341.6,
   "prompt_tokens": 3247,
   "output_tokens": 840,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_7j8rpk9mwhg",
   "run_url": "https://llm-speed.com/r/r_7j8rpk9mwhg",
   "date": "2026-10-14",
   "hardware": "RTX 4090 (24GB) + AMD EPYC 7642 48-Core Processor (48c) + 252GB",
   "model": "Qwen3.8-27B-mtp-on",
   "model_size": null,
   "quant": "Q4_K_M",
   "runtime": "llama.cpp",
   "runtime_version": "0.6.0-dev (build 1, commit df18bee)",
   "workload": "concurrent-decode",
   "suite_version": "suite-v1",
   "decode_tps": 87.04,
   "prefill_tps": null,
   "ttft_ms": null,
   "prompt_tokens": 1024,
   "output_tokens": 256,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_7j8rpk9mwhg",
   "run_url": "https://llm-speed.com/r/r_7j8rpk9mwhg",
   "date": "2026-10-14",
   "hardware": "RTX 4090 (24GB) + AMD EPYC 7642 48-Core Processor (48c) + 252GB",
   "model": "Qwen3.8-27B-mtp-on",
   "model_size": null,
   "quant": "Q4_K_M",
   "runtime": "llama.cpp",
   "runtime_version": "0.6.0-dev (build 1, commit df18bee)",
   "workload": "chat-short",
   "suite_version": "suite-v1",
   "decode_tps": 84.51,
   "prefill_tps": 299.26,
   "ttft_ms": 514.6,
   "prompt_tokens": 154,
   "output_tokens": 256,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_7j8rpk9mwhg",
   "run_url": "https://llm-speed.com/r/r_7j8rpk9mwhg",
   "date": "2026-10-14",
   "hardware": "RTX 4090 (24GB) + AMD EPYC 7642 48-Core Processor (48c) + 252GB",
   "model": "Qwen3.8-27B-mtp-on",
   "model_size": null,
   "quant": "Q4_K_M",
   "runtime": "llama.cpp",
   "runtime_version": "0.6.0-dev (build 1, commit df18bee)",
   "workload": "chat-long",
   "suite_version": "suite-v1",
   "decode_tps": 73.1,
   "prefill_tps": 1585.82,
   "ttft_ms": 2033.0,
   "prompt_tokens": 3224,
   "output_tokens": 1024,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_7j8rpk9mwhg",
   "run_url": "https://llm-speed.com/r/r_7j8rpk9mwhg",
   "date": "2026-10-14",
   "hardware": "RTX 4090 (24GB) + AMD EPYC 7642 48-Core Processor (48c) + 252GB",
   "model": "Qwen3.8-27B-mtp-on",
   "model_size": null,
   "quant": "Q4_K_M",
   "runtime": "llama.cpp",
   "runtime_version": "0.6.0-dev (build 1, commit df18bee)",
   "workload": "agent-trace",
   "suite_version": "suite-v1",
   "decode_tps": 77.6,
   "prefill_tps": 2912.11,
   "ttft_ms": 787.4,
   "prompt_tokens": 3496,
   "output_tokens": 840,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_2np7abzjobe",
   "run_url": "https://llm-speed.com/r/r_2np7abzjobe",
   "date": "2026-10-14",
   "hardware": "RTX 4090 (24GB) + AMD EPYC 7642 48-Core Processor (48c) + 252GB",
   "model": "llama3.1",
   "model_size": "8.0B",
   "quant": "Q4_K_M",
   "runtime": "ollama",
   "runtime_version": "0.31.1",
   "workload": "concurrent-decode",
   "suite_version": "suite-v1",
   "decode_tps": 135.96,
   "prefill_tps": null,
   "ttft_ms": null,
   "prompt_tokens": 1024,
   "output_tokens": 256,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_2np7abzjobe",
   "run_url": "https://llm-speed.com/r/r_2np7abzjobe",
   "date": "2026-10-14",
   "hardware": "RTX 4090 (24GB) + AMD EPYC 7642 48-Core Processor (48c) + 252GB",
   "model": "llama3.1",
   "model_size": "8.0B",
   "quant": "Q4_K_M",
   "runtime": "ollama",
   "runtime_version": "0.31.1",
   "workload": "chat-short",
   "suite_version": "suite-v1",
   "decode_tps": 137.32,
   "prefill_tps": 269.08,
   "ttft_ms": 408.8,
   "prompt_tokens": 110,
   "output_tokens": 256,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_2np7abzjobe",
   "run_url": "https://llm-speed.com/r/r_2np7abzjobe",
   "date": "2026-10-14",
   "hardware": "RTX 4090 (24GB) + AMD EPYC 7642 48-Core Processor (48c) + 252GB",
   "model": "llama3.1",
   "model_size": "8.0B",
   "quant": "Q4_K_M",
   "runtime": "ollama",
   "runtime_version": "0.31.1",
   "workload": "chat-long",
   "suite_version": "suite-v1",
   "decode_tps": 129.24,
   "prefill_tps": 4652.71,
   "ttft_ms": 674.2,
   "prompt_tokens": 3137,
   "output_tokens": 376,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_2np7abzjobe",
   "run_url": "https://llm-speed.com/r/r_2np7abzjobe",
   "date": "2026-10-14",
   "hardware": "RTX 4090 (24GB) + AMD EPYC 7642 48-Core Processor (48c) + 252GB",
   "model": "llama3.1",
   "model_size": "8.0B",
   "quant": "Q4_K_M",
   "runtime": "ollama",
   "runtime_version": "0.31.1",
   "workload": "agent-trace",
   "suite_version": "suite-v1",
   "decode_tps": 130.53,
   "prefill_tps": 3955.23,
   "ttft_ms": 429.2,
   "prompt_tokens": 3210,
   "output_tokens": 840,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_1k8942tw3zf",
   "run_url": "https://llm-speed.com/r/r_1k8942tw3zf",
   "date": "2026-10-14",
   "hardware": "RTX 4090 (24GB) + AMD EPYC 7642 48-Core Processor (48c) + 252GB",
   "model": "Qwen3-Coder-30B-A3B-Instruct",
   "model_size": null,
   "quant": "Q4_K_M",
   "runtime": "llama.cpp",
   "runtime_version": "0.6.0-dev (build 1, commit df18bee)",
   "workload": "concurrent-decode",
   "suite_version": "suite-v1",
   "decode_tps": 225.12,
   "prefill_tps": null,
   "ttft_ms": null,
   "prompt_tokens": 1024,
   "output_tokens": 256,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_1k8942tw3zf",
   "run_url": "https://llm-speed.com/r/r_1k8942tw3zf",
   "date": "2026-10-14",
   "hardware": "RTX 4090 (24GB) + AMD EPYC 7642 48-Core Processor (48c) + 252GB",
   "model": "Qwen3-Coder-30B-A3B-Instruct",
   "model_size": null,
   "quant": "Q4_K_M",
   "runtime": "llama.cpp",
   "runtime_version": "0.6.0-dev (build 1, commit df18bee)",
   "workload": "chat-short",
   "suite_version": "suite-v1",
   "decode_tps": 232.46,
   "prefill_tps": 831.89,
   "ttft_ms": 132.2,
   "prompt_tokens": 110,
   "output_tokens": 256,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_1k8942tw3zf",
   "run_url": "https://llm-speed.com/r/r_1k8942tw3zf",
   "date": "2026-10-14",
   "hardware": "RTX 4090 (24GB) + AMD EPYC 7642 48-Core Processor (48c) + 252GB",
   "model": "Qwen3-Coder-30B-A3B-Instruct",
   "model_size": null,
   "quant": "Q4_K_M",
   "runtime": "llama.cpp",
   "runtime_version": "0.6.0-dev (build 1, commit df18bee)",
   "workload": "chat-long",
   "suite_version": "suite-v1",
   "decode_tps": 206.93,
   "prefill_tps": 7688.67,
   "ttft_ms": 409.3,
   "prompt_tokens": 3147,
   "output_tokens": 561,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_1k8942tw3zf",
   "run_url": "https://llm-speed.com/r/r_1k8942tw3zf",
   "date": "2026-10-14",
   "hardware": "RTX 4090 (24GB) + AMD EPYC 7642 48-Core Processor (48c) + 252GB",
   "model": "Qwen3-Coder-30B-A3B-Instruct",
   "model_size": null,
   "quant": "Q4_K_M",
   "runtime": "llama.cpp",
   "runtime_version": "0.6.0-dev (build 1, commit df18bee)",
   "workload": "agent-trace",
   "suite_version": "suite-v1",
   "decode_tps": 216.35,
   "prefill_tps": 19138.09,
   "ttft_ms": 108.4,
   "prompt_tokens": 3247,
   "output_tokens": 840,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_11kn1epxewh",
   "run_url": "https://llm-speed.com/r/r_11kn1epxewh",
   "date": "2026-10-14",
   "hardware": "RTX 4090 (24GB) + AMD EPYC 7642 48-Core Processor (48c) + 252GB",
   "model": "Qwen3.8-27B",
   "model_size": null,
   "quant": "Q4_K_M",
   "runtime": "llama.cpp",
   "runtime_version": "0.6.0-dev (build 1, commit df18bee)",
   "workload": "concurrent-decode",
   "suite_version": "suite-v1",
   "decode_tps": 43.0,
   "prefill_tps": null,
   "ttft_ms": null,
   "prompt_tokens": 1024,
   "output_tokens": 256,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_11kn1epxewh",
   "run_url": "https://llm-speed.com/r/r_11kn1epxewh",
   "date": "2026-10-14",
   "hardware": "RTX 4090 (24GB) + AMD EPYC 7642 48-Core Processor (48c) + 252GB",
   "model": "Qwen3.8-27B",
   "model_size": null,
   "quant": "Q4_K_M",
   "runtime": "llama.cpp",
   "runtime_version": "0.6.0-dev (build 1, commit df18bee)",
   "workload": "chat-short",
   "suite_version": "suite-v1",
   "decode_tps": 43.06,
   "prefill_tps": 428.56,
   "ttft_ms": 359.3,
   "prompt_tokens": 154,
   "output_tokens": 256,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_11kn1epxewh",
   "run_url": "https://llm-speed.com/r/r_11kn1epxewh",
   "date": "2026-10-14",
   "hardware": "RTX 4090 (24GB) + AMD EPYC 7642 48-Core Processor (48c) + 252GB",
   "model": "Qwen3.8-27B",
   "model_size": null,
   "quant": "Q4_K_M",
   "runtime": "llama.cpp",
   "runtime_version": "0.6.0-dev (build 1, commit df18bee)",
   "workload": "chat-long",
   "suite_version": "suite-v1",
   "decode_tps": 42.63,
   "prefill_tps": 1915.05,
   "ttft_ms": 1683.5,
   "prompt_tokens": 3224,
   "output_tokens": 1024,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_11kn1epxewh",
   "run_url": "https://llm-speed.com/r/r_11kn1epxewh",
   "date": "2026-10-14",
   "hardware": "RTX 4090 (24GB) + AMD EPYC 7642 48-Core Processor (48c) + 252GB",
   "model": "Qwen3.8-27B",
   "model_size": null,
   "quant": "Q4_K_M",
   "runtime": "llama.cpp",
   "runtime_version": "0.6.0-dev (build 1, commit df18bee)",
   "workload": "agent-trace",
   "suite_version": "suite-v1",
   "decode_tps": 42.07,
   "prefill_tps": 3643.98,
   "ttft_ms": 629.3,
   "prompt_tokens": 3496,
   "output_tokens": 840,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_b948ehgi8g4",
   "run_url": "https://llm-speed.com/r/r_b948ehgi8g4",
   "date": "2026-09-19",
   "hardware": "RTX 5880 (48GB) + AMD EPYC 9T24 96-Core Processor (16c) + 62GB",
   "model": "llama3.2",
   "model_size": "1.2B",
   "quant": "Q8_0",
   "runtime": "ollama",
   "runtime_version": "0.34.2",
   "workload": "concurrent-decode",
   "suite_version": "suite-v1",
   "decode_tps": 464.16,
   "prefill_tps": null,
   "ttft_ms": null,
   "prompt_tokens": 1024,
   "output_tokens": 256,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_b948ehgi8g4",
   "run_url": "https://llm-speed.com/r/r_b948ehgi8g4",
   "date": "2026-09-19",
   "hardware": "RTX 5880 (48GB) + AMD EPYC 9T24 96-Core Processor (16c) + 62GB",
   "model": "llama3.2",
   "model_size": "1.2B",
   "quant": "Q8_0",
   "runtime": "ollama",
   "runtime_version": "0.34.2",
   "workload": "chat-short",
   "suite_version": "suite-v1",
   "decode_tps": 466.95,
   "prefill_tps": 44.32,
   "ttft_ms": 2820.6,
   "prompt_tokens": 125,
   "output_tokens": 256,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_b948ehgi8g4",
   "run_url": "https://llm-speed.com/r/r_b948ehgi8g4",
   "date": "2026-09-19",
   "hardware": "RTX 5880 (48GB) + AMD EPYC 9T24 96-Core Processor (16c) + 62GB",
   "model": "llama3.2",
   "model_size": "1.2B",
   "quant": "Q8_0",
   "runtime": "ollama",
   "runtime_version": "0.34.2",
   "workload": "chat-long",
   "suite_version": "suite-v1",
   "decode_tps": 441.9,
   "prefill_tps": 39673.01,
   "ttft_ms": 79.4,
   "prompt_tokens": 3152,
   "output_tokens": 273,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_b948ehgi8g4",
   "run_url": "https://llm-speed.com/r/r_b948ehgi8g4",
   "date": "2026-09-19",
   "hardware": "RTX 5880 (48GB) + AMD EPYC 9T24 96-Core Processor (16c) + 62GB",
   "model": "llama3.2",
   "model_size": "1.2B",
   "quant": "Q8_0",
   "runtime": "ollama",
   "runtime_version": "0.34.2",
   "workload": "agent-trace",
   "suite_version": "suite-v1",
   "decode_tps": 452.31,
   "prefill_tps": 91077.23,
   "ttft_ms": 21.8,
   "prompt_tokens": 3225,
   "output_tokens": 840,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_b1rn5gi6b3b",
   "run_url": "https://llm-speed.com/r/r_b1rn5gi6b3b",
   "date": "2026-09-13",
   "hardware": "AMD64 Family 23 Model 104 Stepping 1, AuthenticAMD (6c) + 15GB",
   "model": "llama3.2",
   "model_size": "1.2B",
   "quant": "Q8_0",
   "runtime": "ollama",
   "runtime_version": "0.34.0",
   "workload": "concurrent-decode",
   "suite_version": "suite-v1",
   "decode_tps": 19.96,
   "prefill_tps": null,
   "ttft_ms": null,
   "prompt_tokens": 1024,
   "output_tokens": 256,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_b1rn5gi6b3b",
   "run_url": "https://llm-speed.com/r/r_b1rn5gi6b3b",
   "date": "2026-09-13",
   "hardware": "AMD64 Family 23 Model 104 Stepping 1, AuthenticAMD (6c) + 15GB",
   "model": "llama3.2",
   "model_size": "1.2B",
   "quant": "Q8_0",
   "runtime": "ollama",
   "runtime_version": "0.34.0",
   "workload": "chat-short",
   "suite_version": "suite-v1",
   "decode_tps": 23.88,
   "prefill_tps": 39.41,
   "ttft_ms": 3171.6,
   "prompt_tokens": 125,
   "output_tokens": 256,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_b1rn5gi6b3b",
   "run_url": "https://llm-speed.com/r/r_b1rn5gi6b3b",
   "date": "2026-09-13",
   "hardware": "AMD64 Family 23 Model 104 Stepping 1, AuthenticAMD (6c) + 15GB",
   "model": "llama3.2",
   "model_size": "1.2B",
   "quant": "Q8_0",
   "runtime": "ollama",
   "runtime_version": "0.34.0",
   "workload": "chat-long",
   "suite_version": "suite-v1",
   "decode_tps": 14.69,
   "prefill_tps": 93.91,
   "ttft_ms": 33563.2,
   "prompt_tokens": 3152,
   "output_tokens": 464,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_b1rn5gi6b3b",
   "run_url": "https://llm-speed.com/r/r_b1rn5gi6b3b",
   "date": "2026-09-13",
   "hardware": "AMD64 Family 23 Model 104 Stepping 1, AuthenticAMD (6c) + 15GB",
   "model": "llama3.2",
   "model_size": "1.2B",
   "quant": "Q8_0",
   "runtime": "ollama",
   "runtime_version": "0.34.0",
   "workload": "agent-trace",
   "suite_version": "suite-v1",
   "decode_tps": 16.8,
   "prefill_tps": 203.16,
   "ttft_ms": 7976.6,
   "prompt_tokens": 3225,
   "output_tokens": 840,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_8q-uf-rq-gp",
   "run_url": "https://llm-speed.com/r/r_8q-uf-rq-gp",
   "date": "2026-09-06",
   "hardware": "RTX 5090 (32GB) + AMD Ryzen 7 9850X3D 8-Core Processor (8c) + 30GB",
   "model": "gemma-4-12b-it-qat",
   "model_size": null,
   "quant": "Q4_0",
   "runtime": "llama.cpp",
   "runtime_version": "1 (9725a31)",
   "workload": "long-context-decay",
   "suite_version": "suite-v1",
   "decode_tps": 134.18,
   "prefill_tps": 6022.36,
   "ttft_ms": 3963.9,
   "prompt_tokens": 23872,
   "output_tokens": 98,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_1srqmpjnkgd",
   "run_url": "https://llm-speed.com/r/r_1srqmpjnkgd",
   "date": "2026-09-06",
   "hardware": "RTX 5090 (32GB) + AMD Ryzen 7 9850X3D 8-Core Processor (8c) + 30GB",
   "model": "gemma-4-12b-it-qat",
   "model_size": null,
   "quant": "Q4_0",
   "runtime": "llama.cpp",
   "runtime_version": "1 (9725a31)",
   "workload": "long-context-decay",
   "suite_version": "suite-v1",
   "decode_tps": 136.27,
   "prefill_tps": 7161.04,
   "ttft_ms": 3333.6,
   "prompt_tokens": 23872,
   "output_tokens": 98,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_-lh-vczf7ik",
   "run_url": "https://llm-speed.com/r/r_-lh-vczf7ik",
   "date": "2026-09-06",
   "hardware": "RTX 5090 (32GB) + AMD Ryzen 7 9850X3D 8-Core Processor (8c) + 30GB",
   "model": "gemma-4-12b-it-qat",
   "model_size": null,
   "quant": "Q4_0",
   "runtime": "llama.cpp",
   "runtime_version": "1 (9725a31)",
   "workload": "long-context-decay",
   "suite_version": "suite-v1",
   "decode_tps": 142.61,
   "prefill_tps": 7115.79,
   "ttft_ms": 3354.8,
   "prompt_tokens": 23872,
   "output_tokens": 98,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_v5arqmazf31",
   "run_url": "https://llm-speed.com/r/r_v5arqmazf31",
   "date": "2026-09-05",
   "hardware": "RTX 5090 (32GB) + AMD Ryzen 7 9850X3D 8-Core Processor (8c) + 30GB",
   "model": "gemma-4-12b-it-qat",
   "model_size": null,
   "quant": "Q4_0",
   "runtime": "llama.cpp",
   "runtime_version": "1 (9725a31)",
   "workload": "chat-short",
   "suite_version": "suite-v1",
   "decode_tps": 141.12,
   "prefill_tps": 2027.92,
   "ttft_ms": 59.7,
   "prompt_tokens": 121,
   "output_tokens": 256,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_v5arqmazf31",
   "run_url": "https://llm-speed.com/r/r_v5arqmazf31",
   "date": "2026-09-05",
   "hardware": "RTX 5090 (32GB) + AMD Ryzen 7 9850X3D 8-Core Processor (8c) + 30GB",
   "model": "gemma-4-12b-it-qat",
   "model_size": null,
   "quant": "Q4_0",
   "runtime": "llama.cpp",
   "runtime_version": "1 (9725a31)",
   "workload": "chat-long",
   "suite_version": "suite-v1",
   "decode_tps": 135.64,
   "prefill_tps": 5147.66,
   "ttft_ms": 619.3,
   "prompt_tokens": 3188,
   "output_tokens": 620,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_dz5a3lzxhrj",
   "run_url": "https://llm-speed.com/r/r_dz5a3lzxhrj",
   "date": "2026-09-05",
   "hardware": "RTX 5090 (32GB) + AMD Ryzen 7 9850X3D 8-Core Processor (8c) + 30GB",
   "model": "Qwen3.8-27B",
   "model_size": null,
   "quant": "Q4_K_M",
   "runtime": "llama.cpp",
   "runtime_version": "1 (9725a31)",
   "workload": "chat-short",
   "suite_version": "suite-v1",
   "decode_tps": 65.66,
   "prefill_tps": 969.83,
   "ttft_ms": 117.5,
   "prompt_tokens": 114,
   "output_tokens": 256,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_dz5a3lzxhrj",
   "run_url": "https://llm-speed.com/r/r_dz5a3lzxhrj",
   "date": "2026-09-05",
   "hardware": "RTX 5090 (32GB) + AMD Ryzen 7 9850X3D 8-Core Processor (8c) + 30GB",
   "model": "Qwen3.8-27B",
   "model_size": null,
   "quant": "Q4_K_M",
   "runtime": "llama.cpp",
   "runtime_version": "1 (9725a31)",
   "workload": "chat-long",
   "suite_version": "suite-v1",
   "decode_tps": 65.1,
   "prefill_tps": 3496.8,
   "ttft_ms": 910.5,
   "prompt_tokens": 3184,
   "output_tokens": 874,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_c3ps9wjygi9",
   "run_url": "https://llm-speed.com/r/r_c3ps9wjygi9",
   "date": "2026-09-05",
   "hardware": "RTX 5090 (32GB) + AMD Ryzen 7 9850X3D 8-Core Processor (8c) + 30GB",
   "model": "Qwen3.8-27B",
   "model_size": null,
   "quant": "Q4_K_M",
   "runtime": "llama.cpp",
   "runtime_version": "1 (9725a31)",
   "workload": "chat-short",
   "suite_version": "suite-v1",
   "decode_tps": 65.73,
   "prefill_tps": 973.57,
   "ttft_ms": 117.1,
   "prompt_tokens": 114,
   "output_tokens": 256,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_c3ps9wjygi9",
   "run_url": "https://llm-speed.com/r/r_c3ps9wjygi9",
   "date": "2026-09-05",
   "hardware": "RTX 5090 (32GB) + AMD Ryzen 7 9850X3D 8-Core Processor (8c) + 30GB",
   "model": "Qwen3.8-27B",
   "model_size": null,
   "quant": "Q4_K_M",
   "runtime": "llama.cpp",
   "runtime_version": "1 (9725a31)",
   "workload": "chat-long",
   "suite_version": "suite-v1",
   "decode_tps": 65.09,
   "prefill_tps": 3506.52,
   "ttft_ms": 908.0,
   "prompt_tokens": 3184,
   "output_tokens": 874,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_c2podhe9qwc",
   "run_url": "https://llm-speed.com/r/r_c2podhe9qwc",
   "date": "2026-09-05",
   "hardware": "RTX 5090 (32GB) + AMD Ryzen 7 9850X3D 8-Core Processor (8c) + 30GB",
   "model": "gemma-4-12b-it-qat",
   "model_size": null,
   "quant": "Q4_0",
   "runtime": "llama.cpp",
   "runtime_version": "1 (9725a31)",
   "workload": "chat-short",
   "suite_version": "suite-v1",
   "decode_tps": 139.5,
   "prefill_tps": 1958.85,
   "ttft_ms": 61.8,
   "prompt_tokens": 121,
   "output_tokens": 256,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_c2podhe9qwc",
   "run_url": "https://llm-speed.com/r/r_c2podhe9qwc",
   "date": "2026-09-05",
   "hardware": "RTX 5090 (32GB) + AMD Ryzen 7 9850X3D 8-Core Processor (8c) + 30GB",
   "model": "gemma-4-12b-it-qat",
   "model_size": null,
   "quant": "Q4_0",
   "runtime": "llama.cpp",
   "runtime_version": "1 (9725a31)",
   "workload": "chat-long",
   "suite_version": "suite-v1",
   "decode_tps": 137.11,
   "prefill_tps": 5117.98,
   "ttft_ms": 622.9,
   "prompt_tokens": 3188,
   "output_tokens": 620,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_8xmm65n1fhq",
   "run_url": "https://llm-speed.com/r/r_8xmm65n1fhq",
   "date": "2026-09-05",
   "hardware": "RTX 5090 (32GB) + AMD Ryzen 7 9850X3D 8-Core Processor (8c) + 30GB",
   "model": "gemma-4-12b-it-qat",
   "model_size": null,
   "quant": "Q4_0",
   "runtime": "llama.cpp",
   "runtime_version": "1 (9725a31)",
   "workload": "chat-short",
   "suite_version": "suite-v1",
   "decode_tps": 137.27,
   "prefill_tps": 2316.39,
   "ttft_ms": 52.2,
   "prompt_tokens": 121,
   "output_tokens": 256,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_8xmm65n1fhq",
   "run_url": "https://llm-speed.com/r/r_8xmm65n1fhq",
   "date": "2026-09-05",
   "hardware": "RTX 5090 (32GB) + AMD Ryzen 7 9850X3D 8-Core Processor (8c) + 30GB",
   "model": "gemma-4-12b-it-qat",
   "model_size": null,
   "quant": "Q4_0",
   "runtime": "llama.cpp",
   "runtime_version": "1 (9725a31)",
   "workload": "chat-long",
   "suite_version": "suite-v1",
   "decode_tps": 135.61,
   "prefill_tps": 5523.3,
   "ttft_ms": 577.2,
   "prompt_tokens": 3188,
   "output_tokens": 620,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_7mod38qsldj",
   "run_url": "https://llm-speed.com/r/r_7mod38qsldj",
   "date": "2026-09-05",
   "hardware": "RTX 5090 (32GB) + AMD Ryzen 7 9850X3D 8-Core Processor (8c) + 30GB",
   "model": "Qwen3.8-27B",
   "model_size": null,
   "quant": "Q4_K_M",
   "runtime": "llama.cpp",
   "runtime_version": "1 (9725a31)",
   "workload": "chat-short",
   "suite_version": "suite-v1",
   "decode_tps": 65.43,
   "prefill_tps": 865.32,
   "ttft_ms": 131.7,
   "prompt_tokens": 114,
   "output_tokens": 256,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_7mod38qsldj",
   "run_url": "https://llm-speed.com/r/r_7mod38qsldj",
   "date": "2026-09-05",
   "hardware": "RTX 5090 (32GB) + AMD Ryzen 7 9850X3D 8-Core Processor (8c) + 30GB",
   "model": "Qwen3.8-27B",
   "model_size": null,
   "quant": "Q4_K_M",
   "runtime": "llama.cpp",
   "runtime_version": "1 (9725a31)",
   "workload": "chat-long",
   "suite_version": "suite-v1",
   "decode_tps": 65.1,
   "prefill_tps": 3501.11,
   "ttft_ms": 909.4,
   "prompt_tokens": 3184,
   "output_tokens": 874,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_w-p7v--eync",
   "run_url": "https://llm-speed.com/r/r_w-p7v--eync",
   "date": "2026-08-18",
   "hardware": "M1 Pro (16-core GPU) + 16GB unified",
   "model": "llama3.2",
   "model_size": "1.2B",
   "quant": "Q8_0",
   "runtime": "ollama",
   "runtime_version": "0.32.14",
   "workload": "concurrent-decode",
   "suite_version": "suite-v1",
   "decode_tps": 97.26,
   "prefill_tps": null,
   "ttft_ms": null,
   "prompt_tokens": 1024,
   "output_tokens": 256,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_w-p7v--eync",
   "run_url": "https://llm-speed.com/r/r_w-p7v--eync",
   "date": "2026-08-18",
   "hardware": "M1 Pro (16-core GPU) + 16GB unified",
   "model": "llama3.2",
   "model_size": "1.2B",
   "quant": "Q8_0",
   "runtime": "ollama",
   "runtime_version": "0.32.14",
   "workload": "chat-short",
   "suite_version": "suite-v1",
   "decode_tps": 103.12,
   "prefill_tps": 8.77,
   "ttft_ms": 14246.7,
   "prompt_tokens": 125,
   "output_tokens": 256,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_w-p7v--eync",
   "run_url": "https://llm-speed.com/r/r_w-p7v--eync",
   "date": "2026-08-18",
   "hardware": "M1 Pro (16-core GPU) + 16GB unified",
   "model": "llama3.2",
   "model_size": "1.2B",
   "quant": "Q8_0",
   "runtime": "ollama",
   "runtime_version": "0.32.14",
   "workload": "chat-long",
   "suite_version": "suite-v1",
   "decode_tps": 81.35,
   "prefill_tps": 1246.52,
   "ttft_ms": 2528.6,
   "prompt_tokens": 3152,
   "output_tokens": 413,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_w-p7v--eync",
   "run_url": "https://llm-speed.com/r/r_w-p7v--eync",
   "date": "2026-08-18",
   "hardware": "M1 Pro (16-core GPU) + 16GB unified",
   "model": "llama3.2",
   "model_size": "1.2B",
   "quant": "Q8_0",
   "runtime": "ollama",
   "runtime_version": "0.32.14",
   "workload": "agent-trace",
   "suite_version": "suite-v1",
   "decode_tps": 93.14,
   "prefill_tps": 2975.32,
   "ttft_ms": 516.3,
   "prompt_tokens": 3225,
   "output_tokens": 840,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_udwgba7udqu",
   "run_url": "https://llm-speed.com/r/r_udwgba7udqu",
   "date": "2026-07-26",
   "hardware": "M5 Max (40-core GPU) + 48GB unified",
   "model": "qwen3.6-agent-q6k-ctx128k",
   "model_size": "35.5B",
   "quant": "Q6_K",
   "runtime": "ollama",
   "runtime_version": "0.32.1",
   "workload": "chat-short",
   "suite_version": "suite-v1",
   "decode_tps": 88.29,
   "prefill_tps": 16.2,
   "ttft_ms": 6915.3,
   "prompt_tokens": 112,
   "output_tokens": 256,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_yy_6jfx70jq",
   "run_url": "https://llm-speed.com/r/r_yy_6jfx70jq",
   "date": "2026-07-12",
   "hardware": "M3 Ultra (60-core GPU) + 96GB unified",
   "model": "lmstudio-community/DeepSeek-R1-0528-Qwen3-8B-MLX-4bit",
   "model_size": null,
   "quant": null,
   "runtime": "mlx",
   "runtime_version": "0.31.3",
   "workload": "concurrent-decode",
   "suite_version": "suite-v1",
   "decode_tps": 110.58,
   "prefill_tps": null,
   "ttft_ms": null,
   "prompt_tokens": 1024,
   "output_tokens": 256,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_yy_6jfx70jq",
   "run_url": "https://llm-speed.com/r/r_yy_6jfx70jq",
   "date": "2026-07-12",
   "hardware": "M3 Ultra (60-core GPU) + 96GB unified",
   "model": "lmstudio-community/DeepSeek-R1-0528-Qwen3-8B-MLX-4bit",
   "model_size": null,
   "quant": null,
   "runtime": "mlx",
   "runtime_version": "0.31.3",
   "workload": "chat-short",
   "suite_version": "suite-v1",
   "decode_tps": 116.82,
   "prefill_tps": 424.47,
   "ttft_ms": 247.4,
   "prompt_tokens": 105,
   "output_tokens": 256,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_yy_6jfx70jq",
   "run_url": "https://llm-speed.com/r/r_yy_6jfx70jq",
   "date": "2026-07-12",
   "hardware": "M3 Ultra (60-core GPU) + 96GB unified",
   "model": "lmstudio-community/DeepSeek-R1-0528-Qwen3-8B-MLX-4bit",
   "model_size": null,
   "quant": null,
   "runtime": "mlx",
   "runtime_version": "0.31.3",
   "workload": "chat-long",
   "suite_version": "suite-v1",
   "decode_tps": 100.49,
   "prefill_tps": 1104.0,
   "ttft_ms": 2846.0,
   "prompt_tokens": 3142,
   "output_tokens": 1024,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_yy_6jfx70jq",
   "run_url": "https://llm-speed.com/r/r_yy_6jfx70jq",
   "date": "2026-07-12",
   "hardware": "M3 Ultra (60-core GPU) + 96GB unified",
   "model": "lmstudio-community/DeepSeek-R1-0528-Qwen3-8B-MLX-4bit",
   "model_size": null,
   "quant": null,
   "runtime": "mlx",
   "runtime_version": "0.31.3",
   "workload": "agent-trace",
   "suite_version": "suite-v1",
   "decode_tps": 107.23,
   "prefill_tps": 1108.91,
   "ttft_ms": 1884.3,
   "prompt_tokens": 3242,
   "output_tokens": 840,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_p2rpb_0iyjj",
   "run_url": "https://llm-speed.com/r/r_p2rpb_0iyjj",
   "date": "2026-07-12",
   "hardware": "M3 Ultra (60-core GPU) + 96GB unified",
   "model": "mlx-community/Qwen3-4B-Instruct-2507-4bit",
   "model_size": null,
   "quant": null,
   "runtime": "mlx",
   "runtime_version": "0.31.3",
   "workload": "concurrent-decode",
   "suite_version": "suite-v1",
   "decode_tps": 168.44,
   "prefill_tps": null,
   "ttft_ms": null,
   "prompt_tokens": 1024,
   "output_tokens": 256,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_p2rpb_0iyjj",
   "run_url": "https://llm-speed.com/r/r_p2rpb_0iyjj",
   "date": "2026-07-12",
   "hardware": "M3 Ultra (60-core GPU) + 96GB unified",
   "model": "mlx-community/Qwen3-4B-Instruct-2507-4bit",
   "model_size": null,
   "quant": null,
   "runtime": "mlx",
   "runtime_version": "0.31.3",
   "workload": "chat-short",
   "suite_version": "suite-v1",
   "decode_tps": 182.68,
   "prefill_tps": 636.62,
   "ttft_ms": 172.8,
   "prompt_tokens": 110,
   "output_tokens": 256,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_p2rpb_0iyjj",
   "run_url": "https://llm-speed.com/r/r_p2rpb_0iyjj",
   "date": "2026-07-12",
   "hardware": "M3 Ultra (60-core GPU) + 96GB unified",
   "model": "mlx-community/Qwen3-4B-Instruct-2507-4bit",
   "model_size": null,
   "quant": null,
   "runtime": "mlx",
   "runtime_version": "0.31.3",
   "workload": "chat-long",
   "suite_version": "suite-v1",
   "decode_tps": 146.98,
   "prefill_tps": 1913.73,
   "ttft_ms": 1644.4,
   "prompt_tokens": 3147,
   "output_tokens": 567,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_p2rpb_0iyjj",
   "run_url": "https://llm-speed.com/r/r_p2rpb_0iyjj",
   "date": "2026-07-12",
   "hardware": "M3 Ultra (60-core GPU) + 96GB unified",
   "model": "mlx-community/Qwen3-4B-Instruct-2507-4bit",
   "model_size": null,
   "quant": null,
   "runtime": "mlx",
   "runtime_version": "0.31.3",
   "workload": "agent-trace",
   "suite_version": "suite-v1",
   "decode_tps": 160.82,
   "prefill_tps": 1913.59,
   "ttft_ms": 1090.9,
   "prompt_tokens": 3247,
   "output_tokens": 840,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_o6wbmfegmde",
   "run_url": "https://llm-speed.com/r/r_o6wbmfegmde",
   "date": "2026-07-12",
   "hardware": "M3 Ultra (60-core GPU) + 96GB unified",
   "model": "mlx-community/Qwen3-30B-A3B-Instruct-2507-4bit",
   "model_size": null,
   "quant": null,
   "runtime": "mlx",
   "runtime_version": "0.31.3",
   "workload": "concurrent-decode",
   "suite_version": "suite-v1",
   "decode_tps": 105.86,
   "prefill_tps": null,
   "ttft_ms": null,
   "prompt_tokens": 1024,
   "output_tokens": 256,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_o6wbmfegmde",
   "run_url": "https://llm-speed.com/r/r_o6wbmfegmde",
   "date": "2026-07-12",
   "hardware": "M3 Ultra (60-core GPU) + 96GB unified",
   "model": "mlx-community/Qwen3-30B-A3B-Instruct-2507-4bit",
   "model_size": null,
   "quant": null,
   "runtime": "mlx",
   "runtime_version": "0.31.3",
   "workload": "chat-short",
   "suite_version": "suite-v1",
   "decode_tps": 112.54,
   "prefill_tps": 308.69,
   "ttft_ms": 356.3,
   "prompt_tokens": 110,
   "output_tokens": 256,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_o6wbmfegmde",
   "run_url": "https://llm-speed.com/r/r_o6wbmfegmde",
   "date": "2026-07-12",
   "hardware": "M3 Ultra (60-core GPU) + 96GB unified",
   "model": "mlx-community/Qwen3-30B-A3B-Instruct-2507-4bit",
   "model_size": null,
   "quant": null,
   "runtime": "mlx",
   "runtime_version": "0.31.3",
   "workload": "chat-long",
   "suite_version": "suite-v1",
   "decode_tps": 96.37,
   "prefill_tps": 1885.17,
   "ttft_ms": 1669.3,
   "prompt_tokens": 3147,
   "output_tokens": 788,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_o6wbmfegmde",
   "run_url": "https://llm-speed.com/r/r_o6wbmfegmde",
   "date": "2026-07-12",
   "hardware": "M3 Ultra (60-core GPU) + 96GB unified",
   "model": "mlx-community/Qwen3-30B-A3B-Instruct-2507-4bit",
   "model_size": null,
   "quant": null,
   "runtime": "mlx",
   "runtime_version": "0.31.3",
   "workload": "agent-trace",
   "suite_version": "suite-v1",
   "decode_tps": 101.26,
   "prefill_tps": 1978.03,
   "ttft_ms": 1057.1,
   "prompt_tokens": 3247,
   "output_tokens": 840,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_b1awwhkoo2v",
   "run_url": "https://llm-speed.com/r/r_b1awwhkoo2v",
   "date": "2026-07-12",
   "hardware": "M3 Ultra (60-core GPU) + 96GB unified",
   "model": "mlx-community/Qwen3-8B-4bit",
   "model_size": null,
   "quant": null,
   "runtime": "mlx",
   "runtime_version": "0.31.3",
   "workload": "concurrent-decode",
   "suite_version": "suite-v1",
   "decode_tps": 114.33,
   "prefill_tps": null,
   "ttft_ms": null,
   "prompt_tokens": 1024,
   "output_tokens": 256,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_b1awwhkoo2v",
   "run_url": "https://llm-speed.com/r/r_b1awwhkoo2v",
   "date": "2026-07-12",
   "hardware": "M3 Ultra (60-core GPU) + 96GB unified",
   "model": "mlx-community/Qwen3-8B-4bit",
   "model_size": null,
   "quant": null,
   "runtime": "mlx",
   "runtime_version": "0.31.3",
   "workload": "chat-short",
   "suite_version": "suite-v1",
   "decode_tps": 120.49,
   "prefill_tps": 446.11,
   "ttft_ms": 246.6,
   "prompt_tokens": 110,
   "output_tokens": 256,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_b1awwhkoo2v",
   "run_url": "https://llm-speed.com/r/r_b1awwhkoo2v",
   "date": "2026-07-12",
   "hardware": "M3 Ultra (60-core GPU) + 96GB unified",
   "model": "mlx-community/Qwen3-8B-4bit",
   "model_size": null,
   "quant": null,
   "runtime": "mlx",
   "runtime_version": "0.31.3",
   "workload": "chat-long",
   "suite_version": "suite-v1",
   "decode_tps": 103.35,
   "prefill_tps": 1108.81,
   "ttft_ms": 2838.2,
   "prompt_tokens": 3147,
   "output_tokens": 1024,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_b1awwhkoo2v",
   "run_url": "https://llm-speed.com/r/r_b1awwhkoo2v",
   "date": "2026-07-12",
   "hardware": "M3 Ultra (60-core GPU) + 96GB unified",
   "model": "mlx-community/Qwen3-8B-4bit",
   "model_size": null,
   "quant": null,
   "runtime": "mlx",
   "runtime_version": "0.31.3",
   "workload": "agent-trace",
   "suite_version": "suite-v1",
   "decode_tps": 109.39,
   "prefill_tps": 1109.15,
   "ttft_ms": 1885.2,
   "prompt_tokens": 3247,
   "output_tokens": 840,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_6rgao6ao735",
   "run_url": "https://llm-speed.com/r/r_6rgao6ao735",
   "date": "2026-07-12",
   "hardware": "M3 Ultra (60-core GPU) + 96GB unified",
   "model": "mlx-community/Qwen3-0.6B-4bit",
   "model_size": null,
   "quant": null,
   "runtime": "mlx",
   "runtime_version": "0.31.3",
   "workload": "concurrent-decode",
   "suite_version": "suite-v1",
   "decode_tps": 368.68,
   "prefill_tps": null,
   "ttft_ms": null,
   "prompt_tokens": 1024,
   "output_tokens": 256,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_6rgao6ao735",
   "run_url": "https://llm-speed.com/r/r_6rgao6ao735",
   "date": "2026-07-12",
   "hardware": "M3 Ultra (60-core GPU) + 96GB unified",
   "model": "mlx-community/Qwen3-0.6B-4bit",
   "model_size": null,
   "quant": null,
   "runtime": "mlx",
   "runtime_version": "0.31.3",
   "workload": "chat-short",
   "suite_version": "suite-v1",
   "decode_tps": 429.23,
   "prefill_tps": 1185.21,
   "ttft_ms": 92.8,
   "prompt_tokens": 110,
   "output_tokens": 256,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_6rgao6ao735",
   "run_url": "https://llm-speed.com/r/r_6rgao6ao735",
   "date": "2026-07-12",
   "hardware": "M3 Ultra (60-core GPU) + 96GB unified",
   "model": "mlx-community/Qwen3-0.6B-4bit",
   "model_size": null,
   "quant": null,
   "runtime": "mlx",
   "runtime_version": "0.31.3",
   "workload": "chat-long",
   "suite_version": "suite-v1",
   "decode_tps": 334.27,
   "prefill_tps": 6893.27,
   "ttft_ms": 456.5,
   "prompt_tokens": 3147,
   "output_tokens": 1024,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_6rgao6ao735",
   "run_url": "https://llm-speed.com/r/r_6rgao6ao735",
   "date": "2026-07-12",
   "hardware": "M3 Ultra (60-core GPU) + 96GB unified",
   "model": "mlx-community/Qwen3-0.6B-4bit",
   "model_size": null,
   "quant": null,
   "runtime": "mlx",
   "runtime_version": "0.31.3",
   "workload": "agent-trace",
   "suite_version": "suite-v1",
   "decode_tps": 368.47,
   "prefill_tps": 8617.74,
   "ttft_ms": 233.1,
   "prompt_tokens": 3247,
   "output_tokens": 840,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_o636l3cc-rr",
   "run_url": "https://llm-speed.com/r/r_o636l3cc-rr",
   "date": "2026-07-08",
   "hardware": "RTX 4090 (24GB) + AMD EPYC 75F3 32-Core Processor (64c) + 504GB",
   "model": "glm-4.7-flash",
   "model_size": "29.9B",
   "quant": "Q4_K_M",
   "runtime": "ollama",
   "runtime_version": "0.31.1",
   "workload": "agent-trace",
   "suite_version": "suite-v1",
   "decode_tps": 129.87,
   "prefill_tps": 2491.13,
   "ttft_ms": 1262.5,
   "prompt_tokens": 3145,
   "output_tokens": 840,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_zrjj-pj93q8",
   "run_url": "https://llm-speed.com/r/r_zrjj-pj93q8",
   "date": "2026-07-02",
   "hardware": "RTX 4090 (24GB) + AMD EPYC 75F3 32-Core Processor (64c) + 504GB",
   "model": "deepseek-r1",
   "model_size": "14.8B",
   "quant": "Q4_K_M",
   "runtime": "ollama",
   "runtime_version": "0.31.1",
   "workload": "chat-long",
   "suite_version": "suite-v1",
   "decode_tps": 81.09,
   "prefill_tps": 240.4,
   "ttft_ms": 13069.6,
   "prompt_tokens": 3142,
   "output_tokens": 1024,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_x23y_sg24pm",
   "run_url": "https://llm-speed.com/r/r_x23y_sg24pm",
   "date": "2026-07-02",
   "hardware": "RTX 4090 (24GB) + AMD EPYC 7443 24-Core Processor (24c) + 252GB",
   "model": "gemma3",
   "model_size": "27.4B",
   "quant": "Q4_K_M",
   "runtime": "ollama",
   "runtime_version": "0.31.1",
   "workload": "concurrent-decode",
   "suite_version": "suite-v1",
   "decode_tps": 46.17,
   "prefill_tps": null,
   "ttft_ms": null,
   "prompt_tokens": 1024,
   "output_tokens": 256,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_x23y_sg24pm",
   "run_url": "https://llm-speed.com/r/r_x23y_sg24pm",
   "date": "2026-07-02",
   "hardware": "RTX 4090 (24GB) + AMD EPYC 7443 24-Core Processor (24c) + 252GB",
   "model": "gemma3",
   "model_size": "27.4B",
   "quant": "Q4_K_M",
   "runtime": "ollama",
   "runtime_version": "0.31.1",
   "workload": "chat-short",
   "suite_version": "suite-v1",
   "decode_tps": 46.95,
   "prefill_tps": 144.51,
   "ttft_ms": 816.6,
   "prompt_tokens": 118,
   "output_tokens": 256,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_x23y_sg24pm",
   "run_url": "https://llm-speed.com/r/r_x23y_sg24pm",
   "date": "2026-07-02",
   "hardware": "RTX 4090 (24GB) + AMD EPYC 7443 24-Core Processor (24c) + 252GB",
   "model": "gemma3",
   "model_size": "27.4B",
   "quant": "Q4_K_M",
   "runtime": "ollama",
   "runtime_version": "0.31.1",
   "workload": "chat-long",
   "suite_version": "suite-v1",
   "decode_tps": 45.25,
   "prefill_tps": 1594.08,
   "ttft_ms": 1998.0,
   "prompt_tokens": 3185,
   "output_tokens": 606,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_x23y_sg24pm",
   "run_url": "https://llm-speed.com/r/r_x23y_sg24pm",
   "date": "2026-07-02",
   "hardware": "RTX 4090 (24GB) + AMD EPYC 7443 24-Core Processor (24c) + 252GB",
   "model": "gemma3",
   "model_size": "27.4B",
   "quant": "Q4_K_M",
   "runtime": "ollama",
   "runtime_version": "0.31.1",
   "workload": "agent-trace",
   "suite_version": "suite-v1",
   "decode_tps": 45.33,
   "prefill_tps": 2556.81,
   "ttft_ms": 955.8,
   "prompt_tokens": 3934,
   "output_tokens": 840,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_wjq32z47vlp",
   "run_url": "https://llm-speed.com/r/r_wjq32z47vlp",
   "date": "2026-07-02",
   "hardware": "RTX 4090 (24GB) + AMD EPYC 7352 24-Core Processor (24c) + 252GB",
   "model": "qwen3-coder",
   "model_size": "30.5B",
   "quant": "Q4_K_M",
   "runtime": "ollama",
   "runtime_version": "0.31.1",
   "workload": "concurrent-decode",
   "suite_version": "suite-v1",
   "decode_tps": 179.89,
   "prefill_tps": null,
   "ttft_ms": null,
   "prompt_tokens": 1024,
   "output_tokens": 256,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_wjq32z47vlp",
   "run_url": "https://llm-speed.com/r/r_wjq32z47vlp",
   "date": "2026-07-02",
   "hardware": "RTX 4090 (24GB) + AMD EPYC 7352 24-Core Processor (24c) + 252GB",
   "model": "qwen3-coder",
   "model_size": "30.5B",
   "quant": "Q4_K_M",
   "runtime": "ollama",
   "runtime_version": "0.31.1",
   "workload": "chat-short",
   "suite_version": "suite-v1",
   "decode_tps": 145.16,
   "prefill_tps": 195.63,
   "ttft_ms": 562.3,
   "prompt_tokens": 110,
   "output_tokens": 256,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_wjq32z47vlp",
   "run_url": "https://llm-speed.com/r/r_wjq32z47vlp",
   "date": "2026-07-02",
   "hardware": "RTX 4090 (24GB) + AMD EPYC 7352 24-Core Processor (24c) + 252GB",
   "model": "qwen3-coder",
   "model_size": "30.5B",
   "quant": "Q4_K_M",
   "runtime": "ollama",
   "runtime_version": "0.31.1",
   "workload": "chat-long",
   "suite_version": "suite-v1",
   "decode_tps": 165.53,
   "prefill_tps": 3527.16,
   "ttft_ms": 892.2,
   "prompt_tokens": 3147,
   "output_tokens": 535,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_pb0jpnbujji",
   "run_url": "https://llm-speed.com/r/r_pb0jpnbujji",
   "date": "2026-07-02",
   "hardware": "RTX 3090 (24GB) + AMD EPYC 7663 56-Core Processor (56c) + 252GB",
   "model": "qwen2.5-coder",
   "model_size": "7.6B",
   "quant": "Q4_K_M",
   "runtime": "ollama",
   "runtime_version": "0.31.1",
   "workload": "concurrent-decode",
   "suite_version": "suite-v1",
   "decode_tps": 139.16,
   "prefill_tps": null,
   "ttft_ms": null,
   "prompt_tokens": 1024,
   "output_tokens": 256,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_pb0jpnbujji",
   "run_url": "https://llm-speed.com/r/r_pb0jpnbujji",
   "date": "2026-07-02",
   "hardware": "RTX 3090 (24GB) + AMD EPYC 7663 56-Core Processor (56c) + 252GB",
   "model": "qwen2.5-coder",
   "model_size": "7.6B",
   "quant": "Q4_K_M",
   "runtime": "ollama",
   "runtime_version": "0.31.1",
   "workload": "chat-short",
   "suite_version": "suite-v1",
   "decode_tps": 39.86,
   "prefill_tps": 2.83,
   "ttft_ms": 46337.3,
   "prompt_tokens": 131,
   "output_tokens": 256,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_pb0jpnbujji",
   "run_url": "https://llm-speed.com/r/r_pb0jpnbujji",
   "date": "2026-07-02",
   "hardware": "RTX 3090 (24GB) + AMD EPYC 7663 56-Core Processor (56c) + 252GB",
   "model": "qwen2.5-coder",
   "model_size": "7.6B",
   "quant": "Q4_K_M",
   "runtime": "ollama",
   "runtime_version": "0.31.1",
   "workload": "chat-long",
   "suite_version": "suite-v1",
   "decode_tps": 135.03,
   "prefill_tps": 3319.44,
   "ttft_ms": 954.4,
   "prompt_tokens": 3168,
   "output_tokens": 412,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_pb0jpnbujji",
   "run_url": "https://llm-speed.com/r/r_pb0jpnbujji",
   "date": "2026-07-02",
   "hardware": "RTX 3090 (24GB) + AMD EPYC 7663 56-Core Processor (56c) + 252GB",
   "model": "qwen2.5-coder",
   "model_size": "7.6B",
   "quant": "Q4_K_M",
   "runtime": "ollama",
   "runtime_version": "0.31.1",
   "workload": "agent-trace",
   "suite_version": "suite-v1",
   "decode_tps": 134.2,
   "prefill_tps": 4111.08,
   "ttft_ms": 486.6,
   "prompt_tokens": 3268,
   "output_tokens": 840,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_o2-1w665rtq",
   "run_url": "https://llm-speed.com/r/r_o2-1w665rtq",
   "date": "2026-07-02",
   "hardware": "RTX 3090 (24GB) + AMD EPYC 7702P 64-Core Processor (64c) + 252GB",
   "model": "deepseek-coder-v2",
   "model_size": "15.7B",
   "quant": "Q4_0",
   "runtime": "ollama",
   "runtime_version": "0.31.1",
   "workload": "concurrent-decode",
   "suite_version": "suite-v1",
   "decode_tps": 155.62,
   "prefill_tps": null,
   "ttft_ms": null,
   "prompt_tokens": 1024,
   "output_tokens": 256,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_o2-1w665rtq",
   "run_url": "https://llm-speed.com/r/r_o2-1w665rtq",
   "date": "2026-07-02",
   "hardware": "RTX 3090 (24GB) + AMD EPYC 7702P 64-Core Processor (64c) + 252GB",
   "model": "deepseek-coder-v2",
   "model_size": "15.7B",
   "quant": "Q4_0",
   "runtime": "ollama",
   "runtime_version": "0.31.1",
   "workload": "chat-short",
   "suite_version": "suite-v1",
   "decode_tps": 189.47,
   "prefill_tps": 159.22,
   "ttft_ms": 734.9,
   "prompt_tokens": 117,
   "output_tokens": 256,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_o2-1w665rtq",
   "run_url": "https://llm-speed.com/r/r_o2-1w665rtq",
   "date": "2026-07-02",
   "hardware": "RTX 3090 (24GB) + AMD EPYC 7702P 64-Core Processor (64c) + 252GB",
   "model": "deepseek-coder-v2",
   "model_size": "15.7B",
   "quant": "Q4_0",
   "runtime": "ollama",
   "runtime_version": "0.31.1",
   "workload": "chat-long",
   "suite_version": "suite-v1",
   "decode_tps": 77.56,
   "prefill_tps": 3907.47,
   "ttft_ms": 838.4,
   "prompt_tokens": 3276,
   "output_tokens": 719,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_o2-1w665rtq",
   "run_url": "https://llm-speed.com/r/r_o2-1w665rtq",
   "date": "2026-07-02",
   "hardware": "RTX 3090 (24GB) + AMD EPYC 7702P 64-Core Processor (64c) + 252GB",
   "model": "deepseek-coder-v2",
   "model_size": "15.7B",
   "quant": "Q4_0",
   "runtime": "ollama",
   "runtime_version": "0.31.1",
   "workload": "agent-trace",
   "suite_version": "suite-v1",
   "decode_tps": 101.15,
   "prefill_tps": 6933.05,
   "ttft_ms": 351.9,
   "prompt_tokens": 4077,
   "output_tokens": 840,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_mv8n8k9wu1e",
   "run_url": "https://llm-speed.com/r/r_mv8n8k9wu1e",
   "date": "2026-07-02",
   "hardware": "RTX 4090 (48GB) + AMD EPYC 7763 64-Core Processor (128c) + 1008GB",
   "model": "qwen2.5-coder",
   "model_size": "7.6B",
   "quant": "Q4_K_M",
   "runtime": "ollama",
   "runtime_version": "0.31.1",
   "workload": "concurrent-decode",
   "suite_version": "suite-v1",
   "decode_tps": 161.15,
   "prefill_tps": null,
   "ttft_ms": null,
   "prompt_tokens": 1024,
   "output_tokens": 256,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_mv8n8k9wu1e",
   "run_url": "https://llm-speed.com/r/r_mv8n8k9wu1e",
   "date": "2026-07-02",
   "hardware": "RTX 4090 (48GB) + AMD EPYC 7763 64-Core Processor (128c) + 1008GB",
   "model": "qwen2.5-coder",
   "model_size": "7.6B",
   "quant": "Q4_K_M",
   "runtime": "ollama",
   "runtime_version": "0.31.1",
   "workload": "chat-short",
   "suite_version": "suite-v1",
   "decode_tps": 159.39,
   "prefill_tps": 451.96,
   "ttft_ms": 289.8,
   "prompt_tokens": 131,
   "output_tokens": 256,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_mv8n8k9wu1e",
   "run_url": "https://llm-speed.com/r/r_mv8n8k9wu1e",
   "date": "2026-07-02",
   "hardware": "RTX 4090 (48GB) + AMD EPYC 7763 64-Core Processor (128c) + 1008GB",
   "model": "qwen2.5-coder",
   "model_size": "7.6B",
   "quant": "Q4_K_M",
   "runtime": "ollama",
   "runtime_version": "0.31.1",
   "workload": "chat-long",
   "suite_version": "suite-v1",
   "decode_tps": 154.73,
   "prefill_tps": 5998.66,
   "ttft_ms": 528.1,
   "prompt_tokens": 3168,
   "output_tokens": 660,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_mv8n8k9wu1e",
   "run_url": "https://llm-speed.com/r/r_mv8n8k9wu1e",
   "date": "2026-07-02",
   "hardware": "RTX 4090 (48GB) + AMD EPYC 7763 64-Core Processor (128c) + 1008GB",
   "model": "qwen2.5-coder",
   "model_size": "7.6B",
   "quant": "Q4_K_M",
   "runtime": "ollama",
   "runtime_version": "0.31.1",
   "workload": "agent-trace",
   "suite_version": "suite-v1",
   "decode_tps": 158.08,
   "prefill_tps": 6421.06,
   "ttft_ms": 320.0,
   "prompt_tokens": 3268,
   "output_tokens": 840,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_l2ck67ls40i",
   "run_url": "https://llm-speed.com/r/r_l2ck67ls40i",
   "date": "2026-07-02",
   "hardware": "RTX 4090 (24GB) + AMD EPYC 75F3 32-Core Processor (64c) + 504GB",
   "model": "deepseek-r1",
   "model_size": "32.8B",
   "quant": "Q4_K_M",
   "runtime": "ollama",
   "runtime_version": "0.31.1",
   "workload": "chat-long",
   "suite_version": "suite-v1",
   "decode_tps": 3.76,
   "prefill_tps": 14.28,
   "ttft_ms": 219986.2,
   "prompt_tokens": 3142,
   "output_tokens": 1024,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_l2ck67ls40i",
   "run_url": "https://llm-speed.com/r/r_l2ck67ls40i",
   "date": "2026-07-02",
   "hardware": "RTX 4090 (24GB) + AMD EPYC 75F3 32-Core Processor (64c) + 504GB",
   "model": "deepseek-r1",
   "model_size": "32.8B",
   "quant": "Q4_K_M",
   "runtime": "ollama",
   "runtime_version": "0.31.1",
   "workload": "agent-trace",
   "suite_version": "suite-v1",
   "decode_tps": 3.79,
   "prefill_tps": 3515.07,
   "ttft_ms": 970.8,
   "prompt_tokens": 3242,
   "output_tokens": 840,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_iu2sfa9ykvw",
   "run_url": "https://llm-speed.com/r/r_iu2sfa9ykvw",
   "date": "2026-07-02",
   "hardware": "RTX 4090 (24GB) + AMD EPYC 7352 24-Core Processor (24c) + 252GB",
   "model": "gpt-oss",
   "model_size": "20.9B",
   "quant": "MXFP4",
   "runtime": "ollama",
   "runtime_version": "0.31.1",
   "workload": "chat-long",
   "suite_version": "suite-v1",
   "decode_tps": 141.74,
   "prefill_tps": 631.51,
   "ttft_ms": 5048.2,
   "prompt_tokens": 3188,
   "output_tokens": 1024,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_h_659oy695r",
   "run_url": "https://llm-speed.com/r/r_h_659oy695r",
   "date": "2026-07-02",
   "hardware": "RTX 4090 (48GB) + AMD EPYC 7763 64-Core Processor (128c) + 1008GB",
   "model": "qwen3.6",
   "model_size": "27.8B",
   "quant": "Q4_K_M",
   "runtime": "ollama",
   "runtime_version": "0.31.1",
   "workload": "agent-trace",
   "suite_version": "suite-v1",
   "decode_tps": 44.18,
   "prefill_tps": 1910.46,
   "ttft_ms": 1807.9,
   "prompt_tokens": 3454,
   "output_tokens": 840,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_h1ub_1uxzdh",
   "run_url": "https://llm-speed.com/r/r_h1ub_1uxzdh",
   "date": "2026-07-02",
   "hardware": "RTX 4090 (48GB) + AMD EPYC 7763 64-Core Processor (128c) + 1008GB",
   "model": "llama3.1",
   "model_size": "8.0B",
   "quant": "Q4_K_M",
   "runtime": "ollama",
   "runtime_version": "0.31.1",
   "workload": "concurrent-decode",
   "suite_version": "suite-v1",
   "decode_tps": 154.23,
   "prefill_tps": null,
   "ttft_ms": null,
   "prompt_tokens": 1024,
   "output_tokens": 256,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_h1ub_1uxzdh",
   "run_url": "https://llm-speed.com/r/r_h1ub_1uxzdh",
   "date": "2026-07-02",
   "hardware": "RTX 4090 (48GB) + AMD EPYC 7763 64-Core Processor (128c) + 1008GB",
   "model": "llama3.1",
   "model_size": "8.0B",
   "quant": "Q4_K_M",
   "runtime": "ollama",
   "runtime_version": "0.31.1",
   "workload": "chat-short",
   "suite_version": "suite-v1",
   "decode_tps": 154.37,
   "prefill_tps": 327.96,
   "ttft_ms": 335.4,
   "prompt_tokens": 110,
   "output_tokens": 256,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_h1ub_1uxzdh",
   "run_url": "https://llm-speed.com/r/r_h1ub_1uxzdh",
   "date": "2026-07-02",
   "hardware": "RTX 4090 (48GB) + AMD EPYC 7763 64-Core Processor (128c) + 1008GB",
   "model": "llama3.1",
   "model_size": "8.0B",
   "quant": "Q4_K_M",
   "runtime": "ollama",
   "runtime_version": "0.31.1",
   "workload": "chat-long",
   "suite_version": "suite-v1",
   "decode_tps": 146.29,
   "prefill_tps": 5032.48,
   "ttft_ms": 623.4,
   "prompt_tokens": 3137,
   "output_tokens": 432,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_h1ub_1uxzdh",
   "run_url": "https://llm-speed.com/r/r_h1ub_1uxzdh",
   "date": "2026-07-02",
   "hardware": "RTX 4090 (48GB) + AMD EPYC 7763 64-Core Processor (128c) + 1008GB",
   "model": "llama3.1",
   "model_size": "8.0B",
   "quant": "Q4_K_M",
   "runtime": "ollama",
   "runtime_version": "0.31.1",
   "workload": "agent-trace",
   "suite_version": "suite-v1",
   "decode_tps": 149.45,
   "prefill_tps": 5377.53,
   "ttft_ms": 387.3,
   "prompt_tokens": 3210,
   "output_tokens": 840,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_fg77v2hhohb",
   "run_url": "https://llm-speed.com/r/r_fg77v2hhohb",
   "date": "2026-07-02",
   "hardware": "RTX 4090 (24GB) + AMD EPYC 75F3 32-Core Processor (64c) + 504GB",
   "model": "deepseek-r1",
   "model_size": "8.2B",
   "quant": "Q4_K_M",
   "runtime": "ollama",
   "runtime_version": "0.31.1",
   "workload": "chat-long",
   "suite_version": "suite-v1",
   "decode_tps": 131.95,
   "prefill_tps": 686.27,
   "ttft_ms": 4578.4,
   "prompt_tokens": 3142,
   "output_tokens": 1024,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_fg77v2hhohb",
   "run_url": "https://llm-speed.com/r/r_fg77v2hhohb",
   "date": "2026-07-02",
   "hardware": "RTX 4090 (24GB) + AMD EPYC 75F3 32-Core Processor (64c) + 504GB",
   "model": "deepseek-r1",
   "model_size": "8.2B",
   "quant": "Q4_K_M",
   "runtime": "ollama",
   "runtime_version": "0.31.1",
   "workload": "agent-trace",
   "suite_version": "suite-v1",
   "decode_tps": 133.84,
   "prefill_tps": 8852.44,
   "ttft_ms": 366.2,
   "prompt_tokens": 3242,
   "output_tokens": 840,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_dlanfbgym0h",
   "run_url": "https://llm-speed.com/r/r_dlanfbgym0h",
   "date": "2026-07-02",
   "hardware": "RTX 4090 (24GB) + AMD EPYC 7443 24-Core Processor (24c) + 252GB",
   "model": "gemma3",
   "model_size": "4.3B",
   "quant": "Q4_K_M",
   "runtime": "ollama",
   "runtime_version": "0.31.1",
   "workload": "concurrent-decode",
   "suite_version": "suite-v1",
   "decode_tps": 193.07,
   "prefill_tps": null,
   "ttft_ms": null,
   "prompt_tokens": 1024,
   "output_tokens": 256,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_dlanfbgym0h",
   "run_url": "https://llm-speed.com/r/r_dlanfbgym0h",
   "date": "2026-07-02",
   "hardware": "RTX 4090 (24GB) + AMD EPYC 7443 24-Core Processor (24c) + 252GB",
   "model": "gemma3",
   "model_size": "4.3B",
   "quant": "Q4_K_M",
   "runtime": "ollama",
   "runtime_version": "0.31.1",
   "workload": "chat-short",
   "suite_version": "suite-v1",
   "decode_tps": 195.0,
   "prefill_tps": 167.24,
   "ttft_ms": 705.6,
   "prompt_tokens": 118,
   "output_tokens": 256,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_dlanfbgym0h",
   "run_url": "https://llm-speed.com/r/r_dlanfbgym0h",
   "date": "2026-07-02",
   "hardware": "RTX 4090 (24GB) + AMD EPYC 7443 24-Core Processor (24c) + 252GB",
   "model": "gemma3",
   "model_size": "4.3B",
   "quant": "Q4_K_M",
   "runtime": "ollama",
   "runtime_version": "0.31.1",
   "workload": "chat-long",
   "suite_version": "suite-v1",
   "decode_tps": 187.56,
   "prefill_tps": 3254.12,
   "ttft_ms": 978.8,
   "prompt_tokens": 3185,
   "output_tokens": 585,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_dlanfbgym0h",
   "run_url": "https://llm-speed.com/r/r_dlanfbgym0h",
   "date": "2026-07-02",
   "hardware": "RTX 4090 (24GB) + AMD EPYC 7443 24-Core Processor (24c) + 252GB",
   "model": "gemma3",
   "model_size": "4.3B",
   "quant": "Q4_K_M",
   "runtime": "ollama",
   "runtime_version": "0.31.1",
   "workload": "agent-trace",
   "suite_version": "suite-v1",
   "decode_tps": 189.99,
   "prefill_tps": 3543.39,
   "ttft_ms": 715.0,
   "prompt_tokens": 3934,
   "output_tokens": 840,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_9hurqggbshk",
   "run_url": "https://llm-speed.com/r/r_9hurqggbshk",
   "date": "2026-07-02",
   "hardware": "RTX 3090 (24GB) + AMD EPYC 7663 56-Core Processor (56c) + 252GB",
   "model": "llama3.1",
   "model_size": "8.0B",
   "quant": "Q4_K_M",
   "runtime": "ollama",
   "runtime_version": "0.31.1",
   "workload": "concurrent-decode",
   "suite_version": "suite-v1",
   "decode_tps": 133.87,
   "prefill_tps": null,
   "ttft_ms": null,
   "prompt_tokens": 1024,
   "output_tokens": 256,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_9hurqggbshk",
   "run_url": "https://llm-speed.com/r/r_9hurqggbshk",
   "date": "2026-07-02",
   "hardware": "RTX 3090 (24GB) + AMD EPYC 7663 56-Core Processor (56c) + 252GB",
   "model": "llama3.1",
   "model_size": "8.0B",
   "quant": "Q4_K_M",
   "runtime": "ollama",
   "runtime_version": "0.31.1",
   "workload": "chat-short",
   "suite_version": "suite-v1",
   "decode_tps": 136.2,
   "prefill_tps": 2.2,
   "ttft_ms": 49960.0,
   "prompt_tokens": 110,
   "output_tokens": 256,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_9hurqggbshk",
   "run_url": "https://llm-speed.com/r/r_9hurqggbshk",
   "date": "2026-07-02",
   "hardware": "RTX 3090 (24GB) + AMD EPYC 7663 56-Core Processor (56c) + 252GB",
   "model": "llama3.1",
   "model_size": "8.0B",
   "quant": "Q4_K_M",
   "runtime": "ollama",
   "runtime_version": "0.31.1",
   "workload": "chat-long",
   "suite_version": "suite-v1",
   "decode_tps": 127.53,
   "prefill_tps": 3004.29,
   "ttft_ms": 1044.2,
   "prompt_tokens": 3137,
   "output_tokens": 390,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_9hurqggbshk",
   "run_url": "https://llm-speed.com/r/r_9hurqggbshk",
   "date": "2026-07-02",
   "hardware": "RTX 3090 (24GB) + AMD EPYC 7663 56-Core Processor (56c) + 252GB",
   "model": "llama3.1",
   "model_size": "8.0B",
   "quant": "Q4_K_M",
   "runtime": "ollama",
   "runtime_version": "0.31.1",
   "workload": "agent-trace",
   "suite_version": "suite-v1",
   "decode_tps": 127.47,
   "prefill_tps": 3475.92,
   "ttft_ms": 527.3,
   "prompt_tokens": 3210,
   "output_tokens": 840,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_73tnfdueq2h",
   "run_url": "https://llm-speed.com/r/r_73tnfdueq2h",
   "date": "2026-07-02",
   "hardware": "RTX 4090 (48GB) + AMD EPYC 7763 64-Core Processor (128c) + 1008GB",
   "model": "qwen2.5-coder",
   "model_size": "14.8B",
   "quant": "Q4_K_M",
   "runtime": "ollama",
   "runtime_version": "0.31.1",
   "workload": "concurrent-decode",
   "suite_version": "suite-v1",
   "decode_tps": 88.63,
   "prefill_tps": null,
   "ttft_ms": null,
   "prompt_tokens": 1024,
   "output_tokens": 256,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_73tnfdueq2h",
   "run_url": "https://llm-speed.com/r/r_73tnfdueq2h",
   "date": "2026-07-02",
   "hardware": "RTX 4090 (48GB) + AMD EPYC 7763 64-Core Processor (128c) + 1008GB",
   "model": "qwen2.5-coder",
   "model_size": "14.8B",
   "quant": "Q4_K_M",
   "runtime": "ollama",
   "runtime_version": "0.31.1",
   "workload": "chat-short",
   "suite_version": "suite-v1",
   "decode_tps": 89.45,
   "prefill_tps": 466.57,
   "ttft_ms": 280.8,
   "prompt_tokens": 131,
   "output_tokens": 256,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_73tnfdueq2h",
   "run_url": "https://llm-speed.com/r/r_73tnfdueq2h",
   "date": "2026-07-02",
   "hardware": "RTX 4090 (48GB) + AMD EPYC 7763 64-Core Processor (128c) + 1008GB",
   "model": "qwen2.5-coder",
   "model_size": "14.8B",
   "quant": "Q4_K_M",
   "runtime": "ollama",
   "runtime_version": "0.31.1",
   "workload": "chat-long",
   "suite_version": "suite-v1",
   "decode_tps": 83.84,
   "prefill_tps": 3572.27,
   "ttft_ms": 886.8,
   "prompt_tokens": 3168,
   "output_tokens": 598,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_73tnfdueq2h",
   "run_url": "https://llm-speed.com/r/r_73tnfdueq2h",
   "date": "2026-07-02",
   "hardware": "RTX 4090 (48GB) + AMD EPYC 7763 64-Core Processor (128c) + 1008GB",
   "model": "qwen2.5-coder",
   "model_size": "14.8B",
   "quant": "Q4_K_M",
   "runtime": "ollama",
   "runtime_version": "0.31.1",
   "workload": "agent-trace",
   "suite_version": "suite-v1",
   "decode_tps": 86.21,
   "prefill_tps": 5298.4,
   "ttft_ms": 394.9,
   "prompt_tokens": 3268,
   "output_tokens": 840,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_6ahy-dq0f_0",
   "run_url": "https://llm-speed.com/r/r_6ahy-dq0f_0",
   "date": "2026-07-02",
   "hardware": "RTX 3090 (24GB) + AMD EPYC 7702P 64-Core Processor (64c) + 252GB",
   "model": "qwen2.5-coder",
   "model_size": "14.8B",
   "quant": "Q4_K_M",
   "runtime": "ollama",
   "runtime_version": "0.31.1",
   "workload": "concurrent-decode",
   "suite_version": "suite-v1",
   "decode_tps": 67.64,
   "prefill_tps": null,
   "ttft_ms": null,
   "prompt_tokens": 1024,
   "output_tokens": 256,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_6ahy-dq0f_0",
   "run_url": "https://llm-speed.com/r/r_6ahy-dq0f_0",
   "date": "2026-07-02",
   "hardware": "RTX 3090 (24GB) + AMD EPYC 7702P 64-Core Processor (64c) + 252GB",
   "model": "qwen2.5-coder",
   "model_size": "14.8B",
   "quant": "Q4_K_M",
   "runtime": "ollama",
   "runtime_version": "0.31.1",
   "workload": "chat-short",
   "suite_version": "suite-v1",
   "decode_tps": 69.21,
   "prefill_tps": 402.05,
   "ttft_ms": 325.8,
   "prompt_tokens": 131,
   "output_tokens": 256,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_6ahy-dq0f_0",
   "run_url": "https://llm-speed.com/r/r_6ahy-dq0f_0",
   "date": "2026-07-02",
   "hardware": "RTX 3090 (24GB) + AMD EPYC 7702P 64-Core Processor (64c) + 252GB",
   "model": "qwen2.5-coder",
   "model_size": "14.8B",
   "quant": "Q4_K_M",
   "runtime": "ollama",
   "runtime_version": "0.31.1",
   "workload": "chat-long",
   "suite_version": "suite-v1",
   "decode_tps": 65.83,
   "prefill_tps": 2070.64,
   "ttft_ms": 1530.0,
   "prompt_tokens": 3168,
   "output_tokens": 532,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_6ahy-dq0f_0",
   "run_url": "https://llm-speed.com/r/r_6ahy-dq0f_0",
   "date": "2026-07-02",
   "hardware": "RTX 3090 (24GB) + AMD EPYC 7702P 64-Core Processor (64c) + 252GB",
   "model": "qwen2.5-coder",
   "model_size": "14.8B",
   "quant": "Q4_K_M",
   "runtime": "ollama",
   "runtime_version": "0.31.1",
   "workload": "agent-trace",
   "suite_version": "suite-v1",
   "decode_tps": 63.63,
   "prefill_tps": 3518.84,
   "ttft_ms": 483.2,
   "prompt_tokens": 3268,
   "output_tokens": 840,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_3kmrc135e0e",
   "run_url": "https://llm-speed.com/r/r_3kmrc135e0e",
   "date": "2026-07-02",
   "hardware": "RTX 4090 (24GB) + AMD EPYC 7443 24-Core Processor (24c) + 252GB",
   "model": "gemma3",
   "model_size": "12.2B",
   "quant": "Q4_K_M",
   "runtime": "ollama",
   "runtime_version": "0.31.1",
   "workload": "concurrent-decode",
   "suite_version": "suite-v1",
   "decode_tps": 90.43,
   "prefill_tps": null,
   "ttft_ms": null,
   "prompt_tokens": 1024,
   "output_tokens": 256,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_3kmrc135e0e",
   "run_url": "https://llm-speed.com/r/r_3kmrc135e0e",
   "date": "2026-07-02",
   "hardware": "RTX 4090 (24GB) + AMD EPYC 7443 24-Core Processor (24c) + 252GB",
   "model": "gemma3",
   "model_size": "12.2B",
   "quant": "Q4_K_M",
   "runtime": "ollama",
   "runtime_version": "0.31.1",
   "workload": "chat-short",
   "suite_version": "suite-v1",
   "decode_tps": 92.65,
   "prefill_tps": 171.36,
   "ttft_ms": 688.6,
   "prompt_tokens": 118,
   "output_tokens": 256,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_3kmrc135e0e",
   "run_url": "https://llm-speed.com/r/r_3kmrc135e0e",
   "date": "2026-07-02",
   "hardware": "RTX 4090 (24GB) + AMD EPYC 7443 24-Core Processor (24c) + 252GB",
   "model": "gemma3",
   "model_size": "12.2B",
   "quant": "Q4_K_M",
   "runtime": "ollama",
   "runtime_version": "0.31.1",
   "workload": "chat-long",
   "suite_version": "suite-v1",
   "decode_tps": 87.6,
   "prefill_tps": 2206.43,
   "ttft_ms": 1443.5,
   "prompt_tokens": 3185,
   "output_tokens": 881,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_3kmrc135e0e",
   "run_url": "https://llm-speed.com/r/r_3kmrc135e0e",
   "date": "2026-07-02",
   "hardware": "RTX 4090 (24GB) + AMD EPYC 7443 24-Core Processor (24c) + 252GB",
   "model": "gemma3",
   "model_size": "12.2B",
   "quant": "Q4_K_M",
   "runtime": "ollama",
   "runtime_version": "0.31.1",
   "workload": "agent-trace",
   "suite_version": "suite-v1",
   "decode_tps": 88.53,
   "prefill_tps": 3123.9,
   "ttft_ms": 815.3,
   "prompt_tokens": 3934,
   "output_tokens": 840,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_2r1p3w9ps7c",
   "run_url": "https://llm-speed.com/r/r_2r1p3w9ps7c",
   "date": "2026-07-02",
   "hardware": "RTX 4090 (24GB) + AMD EPYC 75F3 32-Core Processor (64c) + 504GB",
   "model": "deepseek-r1",
   "model_size": "8.2B",
   "quant": "Q4_K_M",
   "runtime": "ollama",
   "runtime_version": "0.31.1",
   "workload": "chat-long",
   "suite_version": "suite-v1",
   "decode_tps": 130.96,
   "prefill_tps": 599.85,
   "ttft_ms": 5238.0,
   "prompt_tokens": 3142,
   "output_tokens": 1024,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_2r1p3w9ps7c",
   "run_url": "https://llm-speed.com/r/r_2r1p3w9ps7c",
   "date": "2026-07-02",
   "hardware": "RTX 4090 (24GB) + AMD EPYC 75F3 32-Core Processor (64c) + 504GB",
   "model": "deepseek-r1",
   "model_size": "8.2B",
   "quant": "Q4_K_M",
   "runtime": "ollama",
   "runtime_version": "0.31.1",
   "workload": "agent-trace",
   "suite_version": "suite-v1",
   "decode_tps": 132.37,
   "prefill_tps": 9612.78,
   "ttft_ms": 337.3,
   "prompt_tokens": 3242,
   "output_tokens": 840,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_zlh6az5q0o_",
   "run_url": "https://llm-speed.com/r/r_zlh6az5q0o_",
   "date": "2026-07-01",
   "hardware": "RTX 5090 (32GB) + AMD Ryzen 7 9850X3D 8-Core Processor (8c) + 30GB",
   "model": "Yi-Coder-9B-Chat",
   "model_size": null,
   "quant": null,
   "runtime": "llama.cpp",
   "runtime_version": null,
   "workload": "concurrent-decode",
   "suite_version": "suite-v1",
   "decode_tps": 70.99,
   "prefill_tps": null,
   "ttft_ms": null,
   "prompt_tokens": 1024,
   "output_tokens": 256,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_zlh6az5q0o_",
   "run_url": "https://llm-speed.com/r/r_zlh6az5q0o_",
   "date": "2026-07-01",
   "hardware": "RTX 5090 (32GB) + AMD Ryzen 7 9850X3D 8-Core Processor (8c) + 30GB",
   "model": "Yi-Coder-9B-Chat",
   "model_size": null,
   "quant": null,
   "runtime": "llama.cpp",
   "runtime_version": null,
   "workload": "chat-short",
   "suite_version": "suite-v1",
   "decode_tps": 66.79,
   "prefill_tps": null,
   "ttft_ms": 317.0,
   "prompt_tokens": 0,
   "output_tokens": 256,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_zlh6az5q0o_",
   "run_url": "https://llm-speed.com/r/r_zlh6az5q0o_",
   "date": "2026-07-01",
   "hardware": "RTX 5090 (32GB) + AMD Ryzen 7 9850X3D 8-Core Processor (8c) + 30GB",
   "model": "Yi-Coder-9B-Chat",
   "model_size": null,
   "quant": null,
   "runtime": "llama.cpp",
   "runtime_version": null,
   "workload": "chat-long",
   "suite_version": "suite-v1",
   "decode_tps": 69.6,
   "prefill_tps": null,
   "ttft_ms": 1194.6,
   "prompt_tokens": 0,
   "output_tokens": 1024,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_zlh6az5q0o_",
   "run_url": "https://llm-speed.com/r/r_zlh6az5q0o_",
   "date": "2026-07-01",
   "hardware": "RTX 5090 (32GB) + AMD Ryzen 7 9850X3D 8-Core Processor (8c) + 30GB",
   "model": "Yi-Coder-9B-Chat",
   "model_size": null,
   "quant": null,
   "runtime": "llama.cpp",
   "runtime_version": null,
   "workload": "agent-trace",
   "suite_version": "suite-v1",
   "decode_tps": 71.76,
   "prefill_tps": 4478.03,
   "ttft_ms": 397.2,
   "prompt_tokens": 3362,
   "output_tokens": 840,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_yz3q11pa049",
   "run_url": "https://llm-speed.com/r/r_yz3q11pa049",
   "date": "2026-07-01",
   "hardware": "RTX 5090 (32GB) + AMD Ryzen 7 9850X3D 8-Core Processor (8c) + 30GB",
   "model": "gemma-4-31B-it-Q4_K_M.gguf",
   "model_size": null,
   "quant": null,
   "runtime": "llama.cpp",
   "runtime_version": null,
   "workload": "chat-short",
   "suite_version": "suite-v1",
   "decode_tps": 67.7,
   "prefill_tps": null,
   "ttft_ms": 156.4,
   "prompt_tokens": 0,
   "output_tokens": 253,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_yluotk909p8",
   "run_url": "https://llm-speed.com/r/r_yluotk909p8",
   "date": "2026-07-01",
   "hardware": "RTX 5090 (32GB) + AMD Ryzen 7 9850X3D 8-Core Processor (8c) + 30GB",
   "model": "Qwen3.6-27B-Q4_K_M.gguf",
   "model_size": null,
   "quant": null,
   "runtime": "llama.cpp",
   "runtime_version": null,
   "workload": "chat-short",
   "suite_version": "suite-v1",
   "decode_tps": 72.38,
   "prefill_tps": null,
   "ttft_ms": 167.1,
   "prompt_tokens": 0,
   "output_tokens": 256,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_xsajby03vhb",
   "run_url": "https://llm-speed.com/r/r_xsajby03vhb",
   "date": "2026-07-01",
   "hardware": "RTX 5090 (32GB) + AMD Ryzen 7 9850X3D 8-Core Processor (8c) + 30GB",
   "model": "Qwen3.6-35B-A3B-Q4_K_M.gguf",
   "model_size": null,
   "quant": null,
   "runtime": "llama.cpp",
   "runtime_version": null,
   "workload": "chat-short",
   "suite_version": "suite-v1",
   "decode_tps": 222.38,
   "prefill_tps": null,
   "ttft_ms": 294.0,
   "prompt_tokens": 0,
   "output_tokens": 256,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_xr4qdv1hgf2",
   "run_url": "https://llm-speed.com/r/r_xr4qdv1hgf2",
   "date": "2026-07-01",
   "hardware": "RTX 5090 (32GB) + AMD Ryzen 7 9850X3D 8-Core Processor (8c) + 30GB",
   "model": "Qwen2.5-14B-Instruct",
   "model_size": null,
   "quant": null,
   "runtime": "llama.cpp",
   "runtime_version": null,
   "workload": "concurrent-decode",
   "suite_version": "suite-v1",
   "decode_tps": 131.45,
   "prefill_tps": null,
   "ttft_ms": null,
   "prompt_tokens": 1024,
   "output_tokens": 256,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_xr4qdv1hgf2",
   "run_url": "https://llm-speed.com/r/r_xr4qdv1hgf2",
   "date": "2026-07-01",
   "hardware": "RTX 5090 (32GB) + AMD Ryzen 7 9850X3D 8-Core Processor (8c) + 30GB",
   "model": "Qwen2.5-14B-Instruct",
   "model_size": null,
   "quant": null,
   "runtime": "llama.cpp",
   "runtime_version": null,
   "workload": "chat-short",
   "suite_version": "suite-v1",
   "decode_tps": 130.27,
   "prefill_tps": null,
   "ttft_ms": 48.6,
   "prompt_tokens": 0,
   "output_tokens": 256,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_xr4qdv1hgf2",
   "run_url": "https://llm-speed.com/r/r_xr4qdv1hgf2",
   "date": "2026-07-01",
   "hardware": "RTX 5090 (32GB) + AMD Ryzen 7 9850X3D 8-Core Processor (8c) + 30GB",
   "model": "Qwen2.5-14B-Instruct",
   "model_size": null,
   "quant": null,
   "runtime": "llama.cpp",
   "runtime_version": null,
   "workload": "chat-long",
   "suite_version": "suite-v1",
   "decode_tps": 122.71,
   "prefill_tps": null,
   "ttft_ms": 488.8,
   "prompt_tokens": 0,
   "output_tokens": 401,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_xr4qdv1hgf2",
   "run_url": "https://llm-speed.com/r/r_xr4qdv1hgf2",
   "date": "2026-07-01",
   "hardware": "RTX 5090 (32GB) + AMD Ryzen 7 9850X3D 8-Core Processor (8c) + 30GB",
   "model": "Qwen2.5-14B-Instruct",
   "model_size": null,
   "quant": null,
   "runtime": "llama.cpp",
   "runtime_version": null,
   "workload": "agent-trace",
   "suite_version": "suite-v1",
   "decode_tps": 122.77,
   "prefill_tps": 20315.38,
   "ttft_ms": 110.8,
   "prompt_tokens": 3362,
   "output_tokens": 840,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_xbis0gb1uq3",
   "run_url": "https://llm-speed.com/r/r_xbis0gb1uq3",
   "date": "2026-07-01",
   "hardware": "RTX 5090 (32GB) + AMD Ryzen 7 9850X3D 8-Core Processor (8c) + 30GB",
   "model": "Qwen3.6-35B-A3B-Q4_K_M.gguf",
   "model_size": null,
   "quant": null,
   "runtime": "llama.cpp",
   "runtime_version": null,
   "workload": "chat-short",
   "suite_version": "suite-v1",
   "decode_tps": 222.39,
   "prefill_tps": null,
   "ttft_ms": 172.4,
   "prompt_tokens": 0,
   "output_tokens": 256,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_xa08qutcmkm",
   "run_url": "https://llm-speed.com/r/r_xa08qutcmkm",
   "date": "2026-07-01",
   "hardware": "RTX 5090 (32GB) + AMD Ryzen 7 9850X3D 8-Core Processor (8c) + 30GB",
   "model": "Qwen3.6-27B-Q4_K_M.gguf",
   "model_size": null,
   "quant": null,
   "runtime": "llama.cpp",
   "runtime_version": null,
   "workload": "chat-short",
   "suite_version": "suite-v1",
   "decode_tps": 68.93,
   "prefill_tps": null,
   "ttft_ms": 326.4,
   "prompt_tokens": 0,
   "output_tokens": 256,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_wwq2l7lmy6d",
   "run_url": "https://llm-speed.com/r/r_wwq2l7lmy6d",
   "date": "2026-07-01",
   "hardware": "RTX 5090 (32GB) + AMD Ryzen 7 9850X3D 8-Core Processor (8c) + 30GB",
   "model": "gemma-4-31B-it-Q4_K_M.gguf",
   "model_size": null,
   "quant": null,
   "runtime": "llama.cpp",
   "runtime_version": null,
   "workload": "chat-short",
   "suite_version": "suite-v1",
   "decode_tps": 67.31,
   "prefill_tps": null,
   "ttft_ms": 160.4,
   "prompt_tokens": 0,
   "output_tokens": 253,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_wtd90zbbusq",
   "run_url": "https://llm-speed.com/r/r_wtd90zbbusq",
   "date": "2026-07-01",
   "hardware": "RTX 5090 (32GB) + AMD Ryzen 7 9850X3D 8-Core Processor (8c) + 30GB",
   "model": "gemma-4-31B-it-Q4_K_M.gguf",
   "model_size": null,
   "quant": null,
   "runtime": "llama.cpp",
   "runtime_version": null,
   "workload": "chat-short",
   "suite_version": "suite-v1",
   "decode_tps": 66.75,
   "prefill_tps": null,
   "ttft_ms": 194.7,
   "prompt_tokens": 0,
   "output_tokens": 253,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_wppiqxyk9iw",
   "run_url": "https://llm-speed.com/r/r_wppiqxyk9iw",
   "date": "2026-07-01",
   "hardware": "RTX 5090 (32GB) + AMD Ryzen 7 9850X3D 8-Core Processor (8c) + 30GB",
   "model": "gemma-4-31B-it-Q4_K_M.gguf",
   "model_size": null,
   "quant": null,
   "runtime": "llama.cpp",
   "runtime_version": null,
   "workload": "chat-short",
   "suite_version": "suite-v1",
   "decode_tps": 66.39,
   "prefill_tps": null,
   "ttft_ms": 210.3,
   "prompt_tokens": 0,
   "output_tokens": 253,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_wowv__0b5s6",
   "run_url": "https://llm-speed.com/r/r_wowv__0b5s6",
   "date": "2026-07-01",
   "hardware": "RTX 5090 (32GB) + AMD Ryzen 7 9850X3D 8-Core Processor (8c) + 30GB",
   "model": "Qwen3.6-27B-Q4_K_M.gguf",
   "model_size": null,
   "quant": null,
   "runtime": "llama.cpp",
   "runtime_version": null,
   "workload": "chat-short",
   "suite_version": "suite-v1",
   "decode_tps": 70.37,
   "prefill_tps": null,
   "ttft_ms": 268.6,
   "prompt_tokens": 0,
   "output_tokens": 256,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_wmexo75dt0o",
   "run_url": "https://llm-speed.com/r/r_wmexo75dt0o",
   "date": "2026-07-01",
   "hardware": "RTX 5090 (32GB) + AMD Ryzen 7 9850X3D 8-Core Processor (8c) + 30GB",
   "model": "Qwen3.6-27B-Q4_K_M.gguf",
   "model_size": null,
   "quant": null,
   "runtime": "llama.cpp",
   "runtime_version": null,
   "workload": "chat-short",
   "suite_version": "suite-v1",
   "decode_tps": 70.04,
   "prefill_tps": null,
   "ttft_ms": 193.6,
   "prompt_tokens": 0,
   "output_tokens": 256,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_wax_x2ryqhk",
   "run_url": "https://llm-speed.com/r/r_wax_x2ryqhk",
   "date": "2026-07-01",
   "hardware": "RTX 5090 (32GB) + AMD Ryzen 7 9850X3D 8-Core Processor (8c) + 30GB",
   "model": "Qwen3.6-27B-Q4_K_M.gguf",
   "model_size": null,
   "quant": null,
   "runtime": "llama.cpp",
   "runtime_version": null,
   "workload": "chat-short",
   "suite_version": "suite-v1",
   "decode_tps": 73.31,
   "prefill_tps": null,
   "ttft_ms": 296.0,
   "prompt_tokens": 0,
   "output_tokens": 256,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_w482rf72v6z",
   "run_url": "https://llm-speed.com/r/r_w482rf72v6z",
   "date": "2026-07-01",
   "hardware": "RTX 5090 (32GB) + AMD Ryzen 7 9850X3D 8-Core Processor (8c) + 30GB",
   "model": "Qwen3.6-27B-Q4_K_M.gguf",
   "model_size": null,
   "quant": null,
   "runtime": "llama.cpp",
   "runtime_version": null,
   "workload": "chat-short",
   "suite_version": "suite-v1",
   "decode_tps": 69.56,
   "prefill_tps": null,
   "ttft_ms": 174.4,
   "prompt_tokens": 0,
   "output_tokens": 256,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_v983y0y3r2u",
   "run_url": "https://llm-speed.com/r/r_v983y0y3r2u",
   "date": "2026-07-01",
   "hardware": "RTX 5090 (32GB) + AMD Ryzen 7 9850X3D 8-Core Processor (8c) + 30GB",
   "model": "Qwen2.5-Coder-32B-Instruct",
   "model_size": null,
   "quant": null,
   "runtime": "llama.cpp",
   "runtime_version": null,
   "workload": "concurrent-decode",
   "suite_version": "suite-v1",
   "decode_tps": 70.96,
   "prefill_tps": null,
   "ttft_ms": null,
   "prompt_tokens": 1024,
   "output_tokens": 256,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_v983y0y3r2u",
   "run_url": "https://llm-speed.com/r/r_v983y0y3r2u",
   "date": "2026-07-01",
   "hardware": "RTX 5090 (32GB) + AMD Ryzen 7 9850X3D 8-Core Processor (8c) + 30GB",
   "model": "Qwen2.5-Coder-32B-Instruct",
   "model_size": null,
   "quant": null,
   "runtime": "llama.cpp",
   "runtime_version": null,
   "workload": "chat-short",
   "suite_version": "suite-v1",
   "decode_tps": 67.41,
   "prefill_tps": null,
   "ttft_ms": 383.5,
   "prompt_tokens": 0,
   "output_tokens": 256,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_v983y0y3r2u",
   "run_url": "https://llm-speed.com/r/r_v983y0y3r2u",
   "date": "2026-07-01",
   "hardware": "RTX 5090 (32GB) + AMD Ryzen 7 9850X3D 8-Core Processor (8c) + 30GB",
   "model": "Qwen2.5-Coder-32B-Instruct",
   "model_size": null,
   "quant": null,
   "runtime": "llama.cpp",
   "runtime_version": null,
   "workload": "chat-long",
   "suite_version": "suite-v1",
   "decode_tps": 69.15,
   "prefill_tps": null,
   "ttft_ms": 1250.6,
   "prompt_tokens": 0,
   "output_tokens": 1024,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_v983y0y3r2u",
   "run_url": "https://llm-speed.com/r/r_v983y0y3r2u",
   "date": "2026-07-01",
   "hardware": "RTX 5090 (32GB) + AMD Ryzen 7 9850X3D 8-Core Processor (8c) + 30GB",
   "model": "Qwen2.5-Coder-32B-Instruct",
   "model_size": null,
   "quant": null,
   "runtime": "llama.cpp",
   "runtime_version": null,
   "workload": "agent-trace",
   "suite_version": "suite-v1",
   "decode_tps": 68.56,
   "prefill_tps": 4677.1,
   "ttft_ms": 414.2,
   "prompt_tokens": 3362,
   "output_tokens": 840,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_v2vrkr4uah1",
   "run_url": "https://llm-speed.com/r/r_v2vrkr4uah1",
   "date": "2026-07-01",
   "hardware": "RTX 5090 (32GB) + AMD Ryzen 7 9850X3D 8-Core Processor (8c) + 30GB",
   "model": "Qwen3.6-27B-Q4_K_M.gguf",
   "model_size": null,
   "quant": null,
   "runtime": "llama.cpp",
   "runtime_version": null,
   "workload": "chat-short",
   "suite_version": "suite-v1",
   "decode_tps": 72.48,
   "prefill_tps": null,
   "ttft_ms": 171.8,
   "prompt_tokens": 0,
   "output_tokens": 256,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_v05axrj3dti",
   "run_url": "https://llm-speed.com/r/r_v05axrj3dti",
   "date": "2026-07-01",
   "hardware": "RTX 5090 (32GB) + AMD Ryzen 7 9850X3D 8-Core Processor (8c) + 30GB",
   "model": "Qwen3.6-27B-Q4_K_M.gguf",
   "model_size": null,
   "quant": null,
   "runtime": "llama.cpp",
   "runtime_version": null,
   "workload": "chat-short",
   "suite_version": "suite-v1",
   "decode_tps": 68.82,
   "prefill_tps": null,
   "ttft_ms": 413.3,
   "prompt_tokens": 0,
   "output_tokens": 256,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_uvy0zplf-sh",
   "run_url": "https://llm-speed.com/r/r_uvy0zplf-sh",
   "date": "2026-07-01",
   "hardware": "RTX 5090 (32GB) + AMD Ryzen 7 9850X3D 8-Core Processor (8c) + 30GB",
   "model": "Qwen3.6-27B-Q4_K_M.gguf",
   "model_size": null,
   "quant": null,
   "runtime": "llama.cpp",
   "runtime_version": null,
   "workload": "chat-short",
   "suite_version": "suite-v1",
   "decode_tps": 72.36,
   "prefill_tps": null,
   "ttft_ms": 168.2,
   "prompt_tokens": 0,
   "output_tokens": 256,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_uut6m_v6ui9",
   "run_url": "https://llm-speed.com/r/r_uut6m_v6ui9",
   "date": "2026-07-01",
   "hardware": "RTX 5090 (32GB) + AMD Ryzen 7 9850X3D 8-Core Processor (8c) + 30GB",
   "model": "gemma-4-31B-it-Q4_K_M.gguf",
   "model_size": null,
   "quant": null,
   "runtime": "llama.cpp",
   "runtime_version": null,
   "workload": "chat-short",
   "suite_version": "suite-v1",
   "decode_tps": 67.37,
   "prefill_tps": null,
   "ttft_ms": 179.8,
   "prompt_tokens": 0,
   "output_tokens": 253,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_uhv-sxt4h97",
   "run_url": "https://llm-speed.com/r/r_uhv-sxt4h97",
   "date": "2026-07-01",
   "hardware": "RTX 5090 (32GB) + AMD Ryzen 7 9850X3D 8-Core Processor (8c) + 30GB",
   "model": "Qwen3.6-27B-Q4_K_M.gguf",
   "model_size": null,
   "quant": null,
   "runtime": "llama.cpp",
   "runtime_version": null,
   "workload": "chat-short",
   "suite_version": "suite-v1",
   "decode_tps": 72.79,
   "prefill_tps": null,
   "ttft_ms": 186.9,
   "prompt_tokens": 0,
   "output_tokens": 256,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_udgtg0c3pjv",
   "run_url": "https://llm-speed.com/r/r_udgtg0c3pjv",
   "date": "2026-07-01",
   "hardware": "RTX 5090 (32GB) + AMD Ryzen 7 9850X3D 8-Core Processor (8c) + 30GB",
   "model": "Qwen3.6-35B-A3B-Q4_K_M.gguf",
   "model_size": null,
   "quant": null,
   "runtime": "llama.cpp",
   "runtime_version": null,
   "workload": "chat-short",
   "suite_version": "suite-v1",
   "decode_tps": 216.97,
   "prefill_tps": null,
   "ttft_ms": 295.4,
   "prompt_tokens": 0,
   "output_tokens": 256,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_u9eqyrsfrgt",
   "run_url": "https://llm-speed.com/r/r_u9eqyrsfrgt",
   "date": "2026-07-01",
   "hardware": "RTX 5090 (32GB) + AMD Ryzen 7 9850X3D 8-Core Processor (8c) + 30GB",
   "model": "Qwen3.6-27B-Q4_K_M.gguf",
   "model_size": null,
   "quant": null,
   "runtime": "llama.cpp",
   "runtime_version": null,
   "workload": "chat-short",
   "suite_version": "suite-v1",
   "decode_tps": 70.95,
   "prefill_tps": null,
   "ttft_ms": 329.5,
   "prompt_tokens": 0,
   "output_tokens": 256,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_u4wa_y_y3vt",
   "run_url": "https://llm-speed.com/r/r_u4wa_y_y3vt",
   "date": "2026-07-01",
   "hardware": "RTX 5090 (32GB) + AMD Ryzen 7 9850X3D 8-Core Processor (8c) + 30GB",
   "model": "Qwen3.6-27B-Q4_K_M.gguf",
   "model_size": null,
   "quant": null,
   "runtime": "llama.cpp",
   "runtime_version": null,
   "workload": "chat-short",
   "suite_version": "suite-v1",
   "decode_tps": 69.56,
   "prefill_tps": null,
   "ttft_ms": 226.9,
   "prompt_tokens": 0,
   "output_tokens": 256,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_u4iojm6-ekg",
   "run_url": "https://llm-speed.com/r/r_u4iojm6-ekg",
   "date": "2026-07-01",
   "hardware": "RTX 5090 (32GB) + AMD Ryzen 7 9850X3D 8-Core Processor (8c) + 30GB",
   "model": "Yi-Coder-9B-Chat",
   "model_size": null,
   "quant": null,
   "runtime": "llama.cpp",
   "runtime_version": null,
   "workload": "chat-short",
   "suite_version": "suite-v1",
   "decode_tps": 199.33,
   "prefill_tps": null,
   "ttft_ms": 36.2,
   "prompt_tokens": 0,
   "output_tokens": 256,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_twfs86tf_xf",
   "run_url": "https://llm-speed.com/r/r_twfs86tf_xf",
   "date": "2026-07-01",
   "hardware": "RTX 5090 (32GB) + AMD Ryzen 7 9850X3D 8-Core Processor (8c) + 30GB",
   "model": "Qwen2.5-32B-Instruct",
   "model_size": null,
   "quant": null,
   "runtime": "llama.cpp",
   "runtime_version": null,
   "workload": "chat-short",
   "suite_version": "suite-v1",
   "decode_tps": 71.9,
   "prefill_tps": null,
   "ttft_ms": 115.2,
   "prompt_tokens": 0,
   "output_tokens": 256,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_tj9bu7gvnvh",
   "run_url": "https://llm-speed.com/r/r_tj9bu7gvnvh",
   "date": "2026-07-01",
   "hardware": "RTX 5090 (32GB) + AMD Ryzen 7 9850X3D 8-Core Processor (8c) + 30GB",
   "model": "Qwen2.5-14B-Instruct",
   "model_size": null,
   "quant": null,
   "runtime": "llama.cpp",
   "runtime_version": null,
   "workload": "chat-short",
   "suite_version": "suite-v1",
   "decode_tps": 133.29,
   "prefill_tps": null,
   "ttft_ms": 55.9,
   "prompt_tokens": 0,
   "output_tokens": 256,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_th_ounxmvpx",
   "run_url": "https://llm-speed.com/r/r_th_ounxmvpx",
   "date": "2026-07-01",
   "hardware": "RTX 5090 (32GB) + AMD Ryzen 7 9850X3D 8-Core Processor (8c) + 30GB",
   "model": "Qwen3.6-27B-Q4_K_M.gguf",
   "model_size": null,
   "quant": null,
   "runtime": "llama.cpp",
   "runtime_version": null,
   "workload": "chat-short",
   "suite_version": "suite-v1",
   "decode_tps": 69.62,
   "prefill_tps": null,
   "ttft_ms": 177.5,
   "prompt_tokens": 0,
   "output_tokens": 256,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_ta1y71vmvhf",
   "run_url": "https://llm-speed.com/r/r_ta1y71vmvhf",
   "date": "2026-07-01",
   "hardware": "RTX 5090 (32GB) + AMD Ryzen 7 9850X3D 8-Core Processor (8c) + 30GB",
   "model": "Qwen3.6-35B-A3B-Q4_K_M.gguf",
   "model_size": null,
   "quant": null,
   "runtime": "llama.cpp",
   "runtime_version": null,
   "workload": "chat-short",
   "suite_version": "suite-v1",
   "decode_tps": 223.16,
   "prefill_tps": null,
   "ttft_ms": 190.9,
   "prompt_tokens": 0,
   "output_tokens": 256,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_sqr8liqh4ii",
   "run_url": "https://llm-speed.com/r/r_sqr8liqh4ii",
   "date": "2026-07-01",
   "hardware": "RTX 5090 (32GB) + AMD Ryzen 7 9850X3D 8-Core Processor (8c) + 30GB",
   "model": "Codestral-22B-v0.1",
   "model_size": null,
   "quant": null,
   "runtime": "llama.cpp",
   "runtime_version": null,
   "workload": "chat-short",
   "suite_version": "suite-v1",
   "decode_tps": 69.95,
   "prefill_tps": null,
   "ttft_ms": 199.4,
   "prompt_tokens": 0,
   "output_tokens": 256,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_sgqjiuomsug",
   "run_url": "https://llm-speed.com/r/r_sgqjiuomsug",
   "date": "2026-07-01",
   "hardware": "RTX 5090 (32GB) + AMD Ryzen 7 9850X3D 8-Core Processor (8c) + 30GB",
   "model": "Qwen3.6-35B-A3B-Q4_K_M.gguf",
   "model_size": null,
   "quant": null,
   "runtime": "llama.cpp",
   "runtime_version": null,
   "workload": "chat-short",
   "suite_version": "suite-v1",
   "decode_tps": 216.29,
   "prefill_tps": null,
   "ttft_ms": 170.4,
   "prompt_tokens": 0,
   "output_tokens": 256,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_sc_ghtmn-ti",
   "run_url": "https://llm-speed.com/r/r_sc_ghtmn-ti",
   "date": "2026-07-01",
   "hardware": "RTX 5090 (32GB) + AMD Ryzen 7 9850X3D 8-Core Processor (8c) + 30GB",
   "model": "Qwen3.6-35B-A3B-Q4_K_M.gguf",
   "model_size": null,
   "quant": null,
   "runtime": "llama.cpp",
   "runtime_version": null,
   "workload": "chat-short",
   "suite_version": "suite-v1",
   "decode_tps": 219.81,
   "prefill_tps": null,
   "ttft_ms": 178.7,
   "prompt_tokens": 0,
   "output_tokens": 256,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_s0-yb8j14ys",
   "run_url": "https://llm-speed.com/r/r_s0-yb8j14ys",
   "date": "2026-07-01",
   "hardware": "RTX 5090 (32GB) + AMD Ryzen 7 9850X3D 8-Core Processor (8c) + 30GB",
   "model": "gemma-4-31B-it-Q4_K_M.gguf",
   "model_size": null,
   "quant": null,
   "runtime": "llama.cpp",
   "runtime_version": null,
   "workload": "chat-short",
   "suite_version": "suite-v1",
   "decode_tps": 66.19,
   "prefill_tps": null,
   "ttft_ms": 200.0,
   "prompt_tokens": 0,
   "output_tokens": 253,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_rrrs_2wn0__",
   "run_url": "https://llm-speed.com/r/r_rrrs_2wn0__",
   "date": "2026-07-01",
   "hardware": "RTX 5090 (32GB) + AMD Ryzen 7 9850X3D 8-Core Processor (8c) + 30GB",
   "model": "Qwen3.6-35B-A3B-Q4_K_M.gguf",
   "model_size": null,
   "quant": null,
   "runtime": "llama.cpp",
   "runtime_version": null,
   "workload": "chat-short",
   "suite_version": "suite-v1",
   "decode_tps": 216.65,
   "prefill_tps": null,
   "ttft_ms": 348.8,
   "prompt_tokens": 0,
   "output_tokens": 256,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_r9h57uts9lr",
   "run_url": "https://llm-speed.com/r/r_r9h57uts9lr",
   "date": "2026-07-01",
   "hardware": "RTX 5090 (32GB) + AMD Ryzen 7 9850X3D 8-Core Processor (8c) + 30GB",
   "model": "gpt-oss-20b",
   "model_size": null,
   "quant": null,
   "runtime": "llama.cpp",
   "runtime_version": null,
   "workload": "chat-short",
   "suite_version": "suite-v1",
   "decode_tps": 69.42,
   "prefill_tps": null,
   "ttft_ms": 332.5,
   "prompt_tokens": 0,
   "output_tokens": 256,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_qsvrisrvp-e",
   "run_url": "https://llm-speed.com/r/r_qsvrisrvp-e",
   "date": "2026-07-01",
   "hardware": "RTX 5090 (32GB) + AMD Ryzen 7 9850X3D 8-Core Processor (8c) + 30GB",
   "model": "Qwen3.6-27B-Q4_K_M.gguf",
   "model_size": null,
   "quant": null,
   "runtime": "llama.cpp",
   "runtime_version": null,
   "workload": "chat-short",
   "suite_version": "suite-v1",
   "decode_tps": 68.94,
   "prefill_tps": null,
   "ttft_ms": 299.7,
   "prompt_tokens": 0,
   "output_tokens": 256,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_qrpvh5hd9km",
   "run_url": "https://llm-speed.com/r/r_qrpvh5hd9km",
   "date": "2026-07-01",
   "hardware": "RTX 5090 (32GB) + AMD Ryzen 7 9850X3D 8-Core Processor (8c) + 30GB",
   "model": "Qwen3.6-27B-Q4_K_M.gguf",
   "model_size": null,
   "quant": null,
   "runtime": "llama.cpp",
   "runtime_version": null,
   "workload": "chat-short",
   "suite_version": "suite-v1",
   "decode_tps": 69.56,
   "prefill_tps": null,
   "ttft_ms": 205.2,
   "prompt_tokens": 0,
   "output_tokens": 256,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_qr4srge34da",
   "run_url": "https://llm-speed.com/r/r_qr4srge34da",
   "date": "2026-07-01",
   "hardware": "RTX 5090 (32GB) + AMD Ryzen 7 9850X3D 8-Core Processor (8c) + 30GB",
   "model": "Llama-3.1-8B-Instruct",
   "model_size": null,
   "quant": null,
   "runtime": "llama.cpp",
   "runtime_version": null,
   "workload": "concurrent-decode",
   "suite_version": "suite-v1",
   "decode_tps": 223.02,
   "prefill_tps": null,
   "ttft_ms": null,
   "prompt_tokens": 1024,
   "output_tokens": 256,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_qr4srge34da",
   "run_url": "https://llm-speed.com/r/r_qr4srge34da",
   "date": "2026-07-01",
   "hardware": "RTX 5090 (32GB) + AMD Ryzen 7 9850X3D 8-Core Processor (8c) + 30GB",
   "model": "Llama-3.1-8B-Instruct",
   "model_size": null,
   "quant": null,
   "runtime": "llama.cpp",
   "runtime_version": null,
   "workload": "chat-short",
   "suite_version": "suite-v1",
   "decode_tps": 229.44,
   "prefill_tps": null,
   "ttft_ms": 35.6,
   "prompt_tokens": 0,
   "output_tokens": 256,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_qr4srge34da",
   "run_url": "https://llm-speed.com/r/r_qr4srge34da",
   "date": "2026-07-01",
   "hardware": "RTX 5090 (32GB) + AMD Ryzen 7 9850X3D 8-Core Processor (8c) + 30GB",
   "model": "Llama-3.1-8B-Instruct",
   "model_size": null,
   "quant": null,
   "runtime": "llama.cpp",
   "runtime_version": null,
   "workload": "chat-long",
   "suite_version": "suite-v1",
   "decode_tps": 211.41,
   "prefill_tps": null,
   "ttft_ms": 265.7,
   "prompt_tokens": 0,
   "output_tokens": 351,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_qr4srge34da",
   "run_url": "https://llm-speed.com/r/r_qr4srge34da",
   "date": "2026-07-01",
   "hardware": "RTX 5090 (32GB) + AMD Ryzen 7 9850X3D 8-Core Processor (8c) + 30GB",
   "model": "Llama-3.1-8B-Instruct",
   "model_size": null,
   "quant": null,
   "runtime": "llama.cpp",
   "runtime_version": null,
   "workload": "agent-trace",
   "suite_version": "suite-v1",
   "decode_tps": 210.26,
   "prefill_tps": 36720.32,
   "ttft_ms": 60.8,
   "prompt_tokens": 3362,
   "output_tokens": 840,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_qlo6j6lnnh3",
   "run_url": "https://llm-speed.com/r/r_qlo6j6lnnh3",
   "date": "2026-07-01",
   "hardware": "RTX 5090 (32GB) + AMD Ryzen 7 9850X3D 8-Core Processor (8c) + 30GB",
   "model": "Qwen3.6-27B-Q4_K_M.gguf",
   "model_size": null,
   "quant": null,
   "runtime": "llama.cpp",
   "runtime_version": null,
   "workload": "chat-short",
   "suite_version": "suite-v1",
   "decode_tps": 73.2,
   "prefill_tps": null,
   "ttft_ms": 290.5,
   "prompt_tokens": 0,
   "output_tokens": 256,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_qlcfyt0mlv6",
   "run_url": "https://llm-speed.com/r/r_qlcfyt0mlv6",
   "date": "2026-07-01",
   "hardware": "RTX 5090 (32GB) + AMD Ryzen 7 9850X3D 8-Core Processor (8c) + 30GB",
   "model": "Qwen3.6-27B-Q4_K_M.gguf",
   "model_size": null,
   "quant": null,
   "runtime": "llama.cpp",
   "runtime_version": null,
   "workload": "chat-short",
   "suite_version": "suite-v1",
   "decode_tps": 72.19,
   "prefill_tps": null,
   "ttft_ms": 235.4,
   "prompt_tokens": 0,
   "output_tokens": 256,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_ql_2f6nzxj_",
   "run_url": "https://llm-speed.com/r/r_ql_2f6nzxj_",
   "date": "2026-07-01",
   "hardware": "RTX 5090 (32GB) + AMD Ryzen 7 9850X3D 8-Core Processor (8c) + 30GB",
   "model": "Qwen3.6-35B-A3B-Q4_K_M.gguf",
   "model_size": null,
   "quant": null,
   "runtime": "llama.cpp",
   "runtime_version": null,
   "workload": "chat-short",
   "suite_version": "suite-v1",
   "decode_tps": 212.64,
   "prefill_tps": null,
   "ttft_ms": 240.2,
   "prompt_tokens": 0,
   "output_tokens": 256,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_qiq8q_cqfk5",
   "run_url": "https://llm-speed.com/r/r_qiq8q_cqfk5",
   "date": "2026-07-01",
   "hardware": "RTX 5090 (32GB) + AMD Ryzen 7 9850X3D 8-Core Processor (8c) + 30GB",
   "model": "Qwen3.6-27B-Q4_K_M.gguf",
   "model_size": null,
   "quant": null,
   "runtime": "llama.cpp",
   "runtime_version": null,
   "workload": "chat-short",
   "suite_version": "suite-v1",
   "decode_tps": 68.0,
   "prefill_tps": null,
   "ttft_ms": 241.0,
   "prompt_tokens": 0,
   "output_tokens": 256,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_qhp37dyi9n1",
   "run_url": "https://llm-speed.com/r/r_qhp37dyi9n1",
   "date": "2026-07-01",
   "hardware": "RTX 5090 (32GB) + AMD Ryzen 7 9850X3D 8-Core Processor (8c) + 30GB",
   "model": "Qwen3.6-27B-Q4_K_M.gguf",
   "model_size": null,
   "quant": null,
   "runtime": "llama.cpp",
   "runtime_version": null,
   "workload": "chat-short",
   "suite_version": "suite-v1",
   "decode_tps": 67.86,
   "prefill_tps": null,
   "ttft_ms": 184.6,
   "prompt_tokens": 0,
   "output_tokens": 256,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_qe0el0pkw93",
   "run_url": "https://llm-speed.com/r/r_qe0el0pkw93",
   "date": "2026-07-01",
   "hardware": "RTX 5090 (32GB) + AMD Ryzen 7 9850X3D 8-Core Processor (8c) + 30GB",
   "model": "Qwen3.6-27B-Q4_K_M.gguf",
   "model_size": null,
   "quant": null,
   "runtime": "llama.cpp",
   "runtime_version": null,
   "workload": "chat-short",
   "suite_version": "suite-v1",
   "decode_tps": 73.26,
   "prefill_tps": null,
   "ttft_ms": 176.1,
   "prompt_tokens": 0,
   "output_tokens": 256,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_qdm3mkehpyx",
   "run_url": "https://llm-speed.com/r/r_qdm3mkehpyx",
   "date": "2026-07-01",
   "hardware": "RTX 5090 (32GB) + AMD Ryzen 7 9850X3D 8-Core Processor (8c) + 30GB",
   "model": "Qwen3.6-27B-Q4_K_M.gguf",
   "model_size": null,
   "quant": null,
   "runtime": "llama.cpp",
   "runtime_version": null,
   "workload": "chat-short",
   "suite_version": "suite-v1",
   "decode_tps": 69.53,
   "prefill_tps": null,
   "ttft_ms": 187.8,
   "prompt_tokens": 0,
   "output_tokens": 256,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_q9f15lz6831",
   "run_url": "https://llm-speed.com/r/r_q9f15lz6831",
   "date": "2026-07-01",
   "hardware": "RTX 5090 (32GB) + AMD Ryzen 7 9850X3D 8-Core Processor (8c) + 30GB",
   "model": "stable-code-instruct-3b",
   "model_size": null,
   "quant": null,
   "runtime": "llama.cpp",
   "runtime_version": null,
   "workload": "concurrent-decode",
   "suite_version": "suite-v1",
   "decode_tps": 356.08,
   "prefill_tps": null,
   "ttft_ms": null,
   "prompt_tokens": 1024,
   "output_tokens": 256,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_q9f15lz6831",
   "run_url": "https://llm-speed.com/r/r_q9f15lz6831",
   "date": "2026-07-01",
   "hardware": "RTX 5090 (32GB) + AMD Ryzen 7 9850X3D 8-Core Processor (8c) + 30GB",
   "model": "stable-code-instruct-3b",
   "model_size": null,
   "quant": null,
   "runtime": "llama.cpp",
   "runtime_version": null,
   "workload": "chat-short",
   "suite_version": "suite-v1",
   "decode_tps": 341.95,
   "prefill_tps": null,
   "ttft_ms": 174.9,
   "prompt_tokens": 0,
   "output_tokens": 256,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_q9f15lz6831",
   "run_url": "https://llm-speed.com/r/r_q9f15lz6831",
   "date": "2026-07-01",
   "hardware": "RTX 5090 (32GB) + AMD Ryzen 7 9850X3D 8-Core Processor (8c) + 30GB",
   "model": "stable-code-instruct-3b",
   "model_size": null,
   "quant": null,
   "runtime": "llama.cpp",
   "runtime_version": null,
   "workload": "chat-long",
   "suite_version": "suite-v1",
   "decode_tps": 289.09,
   "prefill_tps": null,
   "ttft_ms": 231.6,
   "prompt_tokens": 0,
   "output_tokens": 234,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_q9f15lz6831",
   "run_url": "https://llm-speed.com/r/r_q9f15lz6831",
   "date": "2026-07-01",
   "hardware": "RTX 5090 (32GB) + AMD Ryzen 7 9850X3D 8-Core Processor (8c) + 30GB",
   "model": "stable-code-instruct-3b",
   "model_size": null,
   "quant": null,
   "runtime": "llama.cpp",
   "runtime_version": null,
   "workload": "agent-trace",
   "suite_version": "suite-v1",
   "decode_tps": 292.84,
   "prefill_tps": 44387.47,
   "ttft_ms": 42.7,
   "prompt_tokens": 3362,
   "output_tokens": 840,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_q5xsw1p-7f8",
   "run_url": "https://llm-speed.com/r/r_q5xsw1p-7f8",
   "date": "2026-07-01",
   "hardware": "RTX 5090 (32GB) + AMD Ryzen 7 9850X3D 8-Core Processor (8c) + 30GB",
   "model": "Qwen3.6-27B-Q4_K_M.gguf",
   "model_size": null,
   "quant": null,
   "runtime": "llama.cpp",
   "runtime_version": null,
   "workload": "chat-short",
   "suite_version": "suite-v1",
   "decode_tps": 69.95,
   "prefill_tps": null,
   "ttft_ms": 290.7,
   "prompt_tokens": 0,
   "output_tokens": 256,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_q0wcxrit2bz",
   "run_url": "https://llm-speed.com/r/r_q0wcxrit2bz",
   "date": "2026-07-01",
   "hardware": "RTX 5090 (32GB) + AMD Ryzen 7 9850X3D 8-Core Processor (8c) + 30GB",
   "model": "Qwen3.6-27B-Q4_K_M.gguf",
   "model_size": null,
   "quant": null,
   "runtime": "llama.cpp",
   "runtime_version": null,
   "workload": "chat-short",
   "suite_version": "suite-v1",
   "decode_tps": 69.23,
   "prefill_tps": null,
   "ttft_ms": 342.5,
   "prompt_tokens": 0,
   "output_tokens": 256,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_q-it_4s9z4p",
   "run_url": "https://llm-speed.com/r/r_q-it_4s9z4p",
   "date": "2026-07-01",
   "hardware": "RTX 5090 (32GB) + AMD Ryzen 7 9850X3D 8-Core Processor (8c) + 30GB",
   "model": "gemma-4-31B-it-Q4_K_M.gguf",
   "model_size": null,
   "quant": null,
   "runtime": "llama.cpp",
   "runtime_version": null,
   "workload": "chat-short",
   "suite_version": "suite-v1",
   "decode_tps": 67.19,
   "prefill_tps": null,
   "ttft_ms": 230.1,
   "prompt_tokens": 0,
   "output_tokens": 253,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_pnk9rze10h7",
   "run_url": "https://llm-speed.com/r/r_pnk9rze10h7",
   "date": "2026-07-01",
   "hardware": "RTX 5090 (32GB) + AMD Ryzen 7 9850X3D 8-Core Processor (8c) + 30GB",
   "model": "gemma-4-31B-it-Q4_K_M.gguf",
   "model_size": null,
   "quant": null,
   "runtime": "llama.cpp",
   "runtime_version": null,
   "workload": "chat-short",
   "suite_version": "suite-v1",
   "decode_tps": 67.16,
   "prefill_tps": null,
   "ttft_ms": 163.5,
   "prompt_tokens": 0,
   "output_tokens": 253,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_pnei9cgfzl5",
   "run_url": "https://llm-speed.com/r/r_pnei9cgfzl5",
   "date": "2026-07-01",
   "hardware": "RTX 5090 (32GB) + AMD Ryzen 7 9850X3D 8-Core Processor (8c) + 30GB",
   "model": "Qwen3.6-35B-A3B-Q4_K_M.gguf",
   "model_size": null,
   "quant": null,
   "runtime": "llama.cpp",
   "runtime_version": null,
   "workload": "chat-short",
   "suite_version": "suite-v1",
   "decode_tps": 214.04,
   "prefill_tps": null,
   "ttft_ms": 210.6,
   "prompt_tokens": 0,
   "output_tokens": 256,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_pm_a1uf2ufc",
   "run_url": "https://llm-speed.com/r/r_pm_a1uf2ufc",
   "date": "2026-07-01",
   "hardware": "RTX 5090 (32GB) + AMD Ryzen 7 9850X3D 8-Core Processor (8c) + 30GB",
   "model": "Qwen3-Coder-30B-A3B-Instruct",
   "model_size": null,
   "quant": null,
   "runtime": "llama.cpp",
   "runtime_version": null,
   "workload": "concurrent-decode",
   "suite_version": "suite-v1",
   "decode_tps": 68.04,
   "prefill_tps": null,
   "ttft_ms": null,
   "prompt_tokens": 1024,
   "output_tokens": 256,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_pm_a1uf2ufc",
   "run_url": "https://llm-speed.com/r/r_pm_a1uf2ufc",
   "date": "2026-07-01",
   "hardware": "RTX 5090 (32GB) + AMD Ryzen 7 9850X3D 8-Core Processor (8c) + 30GB",
   "model": "Qwen3-Coder-30B-A3B-Instruct",
   "model_size": null,
   "quant": null,
   "runtime": "llama.cpp",
   "runtime_version": null,
   "workload": "chat-short",
   "suite_version": "suite-v1",
   "decode_tps": 68.41,
   "prefill_tps": null,
   "ttft_ms": 333.6,
   "prompt_tokens": 0,
   "output_tokens": 256,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_pm_a1uf2ufc",
   "run_url": "https://llm-speed.com/r/r_pm_a1uf2ufc",
   "date": "2026-07-01",
   "hardware": "RTX 5090 (32GB) + AMD Ryzen 7 9850X3D 8-Core Processor (8c) + 30GB",
   "model": "Qwen3-Coder-30B-A3B-Instruct",
   "model_size": null,
   "quant": null,
   "runtime": "llama.cpp",
   "runtime_version": null,
   "workload": "chat-long",
   "suite_version": "suite-v1",
   "decode_tps": 70.55,
   "prefill_tps": null,
   "ttft_ms": 1203.7,
   "prompt_tokens": 0,
   "output_tokens": 1024,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_pm_a1uf2ufc",
   "run_url": "https://llm-speed.com/r/r_pm_a1uf2ufc",
   "date": "2026-07-01",
   "hardware": "RTX 5090 (32GB) + AMD Ryzen 7 9850X3D 8-Core Processor (8c) + 30GB",
   "model": "Qwen3-Coder-30B-A3B-Instruct",
   "model_size": null,
   "quant": null,
   "runtime": "llama.cpp",
   "runtime_version": null,
   "workload": "agent-trace",
   "suite_version": "suite-v1",
   "decode_tps": 66.09,
   "prefill_tps": 4816.7,
   "ttft_ms": 450.7,
   "prompt_tokens": 3362,
   "output_tokens": 840,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_plujbqnef08",
   "run_url": "https://llm-speed.com/r/r_plujbqnef08",
   "date": "2026-07-01",
   "hardware": "RTX 5090 (32GB) + AMD Ryzen 7 9850X3D 8-Core Processor (8c) + 30GB",
   "model": "gemma-4-31B-it-Q4_K_M.gguf",
   "model_size": null,
   "quant": null,
   "runtime": "llama.cpp",
   "runtime_version": null,
   "workload": "chat-short",
   "suite_version": "suite-v1",
   "decode_tps": 67.37,
   "prefill_tps": null,
   "ttft_ms": 175.8,
   "prompt_tokens": 0,
   "output_tokens": 253,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_phvxm9dcak0",
   "run_url": "https://llm-speed.com/r/r_phvxm9dcak0",
   "date": "2026-07-01",
   "hardware": "RTX 5090 (32GB) + AMD Ryzen 7 9850X3D 8-Core Processor (8c) + 30GB",
   "model": "Qwen3-32B",
   "model_size": null,
   "quant": null,
   "runtime": "llama.cpp",
   "runtime_version": null,
   "workload": "chat-short",
   "suite_version": "suite-v1",
   "decode_tps": 69.45,
   "prefill_tps": null,
   "ttft_ms": 327.1,
   "prompt_tokens": 0,
   "output_tokens": 256,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_pa7nuerd9tl",
   "run_url": "https://llm-speed.com/r/r_pa7nuerd9tl",
   "date": "2026-07-01",
   "hardware": "RTX 5090 (32GB) + AMD Ryzen 7 9850X3D 8-Core Processor (8c) + 30GB",
   "model": "gemma-4-31B-it-Q4_K_M.gguf",
   "model_size": null,
   "quant": null,
   "runtime": "llama.cpp",
   "runtime_version": null,
   "workload": "chat-short",
   "suite_version": "suite-v1",
   "decode_tps": 69.67,
   "prefill_tps": null,
   "ttft_ms": 174.8,
   "prompt_tokens": 0,
   "output_tokens": 253,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_p_f63tcgans",
   "run_url": "https://llm-speed.com/r/r_p_f63tcgans",
   "date": "2026-07-01",
   "hardware": "RTX 5090 (32GB) + AMD Ryzen 7 9850X3D 8-Core Processor (8c) + 30GB",
   "model": "Qwen2.5-Coder-14B-Instruct",
   "model_size": null,
   "quant": null,
   "runtime": "llama.cpp",
   "runtime_version": null,
   "workload": "chat-short",
   "suite_version": "suite-v1",
   "decode_tps": 136.0,
   "prefill_tps": null,
   "ttft_ms": 51.3,
   "prompt_tokens": 0,
   "output_tokens": 256,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_p3m8a1lj4kg",
   "run_url": "https://llm-speed.com/r/r_p3m8a1lj4kg",
   "date": "2026-07-01",
   "hardware": "RTX 5090 (32GB) + AMD Ryzen 7 9850X3D 8-Core Processor (8c) + 30GB",
   "model": "Qwen3.6-27B-Q4_K_M.gguf",
   "model_size": null,
   "quant": null,
   "runtime": "llama.cpp",
   "runtime_version": null,
   "workload": "chat-short",
   "suite_version": "suite-v1",
   "decode_tps": 72.39,
   "prefill_tps": null,
   "ttft_ms": 224.9,
   "prompt_tokens": 0,
   "output_tokens": 256,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_p307g-1cdkl",
   "run_url": "https://llm-speed.com/r/r_p307g-1cdkl",
   "date": "2026-07-01",
   "hardware": "RTX 5090 (32GB) + AMD Ryzen 7 9850X3D 8-Core Processor (8c) + 30GB",
   "model": "Qwen3.6-27B-Q4_K_M.gguf",
   "model_size": null,
   "quant": null,
   "runtime": "llama.cpp",
   "runtime_version": null,
   "workload": "chat-short",
   "suite_version": "suite-v1",
   "decode_tps": 72.79,
   "prefill_tps": null,
   "ttft_ms": 297.2,
   "prompt_tokens": 0,
   "output_tokens": 256,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_p01v4lc4c11",
   "run_url": "https://llm-speed.com/r/r_p01v4lc4c11",
   "date": "2026-07-01",
   "hardware": "RTX 5090 (32GB) + AMD Ryzen 7 9850X3D 8-Core Processor (8c) + 30GB",
   "model": "Qwen3.6-27B-Q4_K_M.gguf",
   "model_size": null,
   "quant": null,
   "runtime": "llama.cpp",
   "runtime_version": null,
   "workload": "chat-short",
   "suite_version": "suite-v1",
   "decode_tps": 69.48,
   "prefill_tps": null,
   "ttft_ms": 251.7,
   "prompt_tokens": 0,
   "output_tokens": 256,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_p-pbvaxim4m",
   "run_url": "https://llm-speed.com/r/r_p-pbvaxim4m",
   "date": "2026-07-01",
   "hardware": "RTX 5090 (32GB) + AMD Ryzen 7 9850X3D 8-Core Processor (8c) + 30GB",
   "model": "Qwen3.6-35B-A3B-Q4_K_M.gguf",
   "model_size": null,
   "quant": null,
   "runtime": "llama.cpp",
   "runtime_version": null,
   "workload": "chat-short",
   "suite_version": "suite-v1",
   "decode_tps": 223.84,
   "prefill_tps": null,
   "ttft_ms": 147.9,
   "prompt_tokens": 0,
   "output_tokens": 256,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_opuj21f13-_",
   "run_url": "https://llm-speed.com/r/r_opuj21f13-_",
   "date": "2026-07-01",
   "hardware": "RTX 5090 (32GB) + AMD Ryzen 7 9850X3D 8-Core Processor (8c) + 30GB",
   "model": "Qwen2.5-Coder-14B-Instruct",
   "model_size": null,
   "quant": null,
   "runtime": "llama.cpp",
   "runtime_version": null,
   "workload": "concurrent-decode",
   "suite_version": "suite-v1",
   "decode_tps": 133.78,
   "prefill_tps": null,
   "ttft_ms": null,
   "prompt_tokens": 1024,
   "output_tokens": 256,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_opuj21f13-_",
   "run_url": "https://llm-speed.com/r/r_opuj21f13-_",
   "date": "2026-07-01",
   "hardware": "RTX 5090 (32GB) + AMD Ryzen 7 9850X3D 8-Core Processor (8c) + 30GB",
   "model": "Qwen2.5-Coder-14B-Instruct",
   "model_size": null,
   "quant": null,
   "runtime": "llama.cpp",
   "runtime_version": null,
   "workload": "chat-short",
   "suite_version": "suite-v1",
   "decode_tps": 140.03,
   "prefill_tps": null,
   "ttft_ms": 42.1,
   "prompt_tokens": 0,
   "output_tokens": 256,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_opuj21f13-_",
   "run_url": "https://llm-speed.com/r/r_opuj21f13-_",
   "date": "2026-07-01",
   "hardware": "RTX 5090 (32GB) + AMD Ryzen 7 9850X3D 8-Core Processor (8c) + 30GB",
   "model": "Qwen2.5-Coder-14B-Instruct",
   "model_size": null,
   "quant": null,
   "runtime": "llama.cpp",
   "runtime_version": null,
   "workload": "chat-long",
   "suite_version": "suite-v1",
   "decode_tps": 122.98,
   "prefill_tps": null,
   "ttft_ms": 457.2,
   "prompt_tokens": 0,
   "output_tokens": 518,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_opuj21f13-_",
   "run_url": "https://llm-speed.com/r/r_opuj21f13-_",
   "date": "2026-07-01",
   "hardware": "RTX 5090 (32GB) + AMD Ryzen 7 9850X3D 8-Core Processor (8c) + 30GB",
   "model": "Qwen2.5-Coder-14B-Instruct",
   "model_size": null,
   "quant": null,
   "runtime": "llama.cpp",
   "runtime_version": null,
   "workload": "agent-trace",
   "suite_version": "suite-v1",
   "decode_tps": 130.45,
   "prefill_tps": 21985.95,
   "ttft_ms": 103.7,
   "prompt_tokens": 3362,
   "output_tokens": 840,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_onvh-dpmck-",
   "run_url": "https://llm-speed.com/r/r_onvh-dpmck-",
   "date": "2026-07-01",
   "hardware": "RTX 5090 (32GB) + AMD Ryzen 7 9850X3D 8-Core Processor (8c) + 30GB",
   "model": "gemma-4-31B-it-Q4_K_M.gguf",
   "model_size": null,
   "quant": null,
   "runtime": "llama.cpp",
   "runtime_version": null,
   "workload": "chat-short",
   "suite_version": "suite-v1",
   "decode_tps": 67.05,
   "prefill_tps": null,
   "ttft_ms": 206.8,
   "prompt_tokens": 0,
   "output_tokens": 253,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_nkbs6d3-d21",
   "run_url": "https://llm-speed.com/r/r_nkbs6d3-d21",
   "date": "2026-07-01",
   "hardware": "RTX 5090 (32GB) + AMD Ryzen 7 9850X3D 8-Core Processor (8c) + 30GB",
   "model": "Qwen2.5-Coder-32B-Instruct",
   "model_size": null,
   "quant": null,
   "runtime": "llama.cpp",
   "runtime_version": null,
   "workload": "chat-short",
   "suite_version": "suite-v1",
   "decode_tps": 71.91,
   "prefill_tps": null,
   "ttft_ms": 71.0,
   "prompt_tokens": 0,
   "output_tokens": 256,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_n6su7ptrulh",
   "run_url": "https://llm-speed.com/r/r_n6su7ptrulh",
   "date": "2026-07-01",
   "hardware": "RTX 5090 (32GB) + AMD Ryzen 7 9850X3D 8-Core Processor (8c) + 30GB",
   "model": "gemma-4-31B-it-Q4_K_M.gguf",
   "model_size": null,
   "quant": null,
   "runtime": "llama.cpp",
   "runtime_version": null,
   "workload": "chat-short",
   "suite_version": "suite-v1",
   "decode_tps": 67.39,
   "prefill_tps": null,
   "ttft_ms": 172.4,
   "prompt_tokens": 0,
   "output_tokens": 253,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_n65ex7zl6ts",
   "run_url": "https://llm-speed.com/r/r_n65ex7zl6ts",
   "date": "2026-07-01",
   "hardware": "RTX 5090 (32GB) + AMD Ryzen 7 9850X3D 8-Core Processor (8c) + 30GB",
   "model": "gemma-4-31B-it-Q4_K_M.gguf",
   "model_size": null,
   "quant": null,
   "runtime": "llama.cpp",
   "runtime_version": null,
   "workload": "chat-short",
   "suite_version": "suite-v1",
   "decode_tps": 69.36,
   "prefill_tps": null,
   "ttft_ms": 145.9,
   "prompt_tokens": 0,
   "output_tokens": 253,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_mqx-i-thqtq",
   "run_url": "https://llm-speed.com/r/r_mqx-i-thqtq",
   "date": "2026-07-01",
   "hardware": "RTX 5090 (32GB) + AMD Ryzen 7 9850X3D 8-Core Processor (8c) + 30GB",
   "model": "Qwen3.6-27B-Q4_K_M.gguf",
   "model_size": null,
   "quant": null,
   "runtime": "llama.cpp",
   "runtime_version": null,
   "workload": "chat-short",
   "suite_version": "suite-v1",
   "decode_tps": 69.65,
   "prefill_tps": null,
   "ttft_ms": 308.2,
   "prompt_tokens": 0,
   "output_tokens": 256,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_moawmhigq5d",
   "run_url": "https://llm-speed.com/r/r_moawmhigq5d",
   "date": "2026-07-01",
   "hardware": "RTX 5090 (32GB) + AMD Ryzen 7 9850X3D 8-Core Processor (8c) + 30GB",
   "model": "gemma-4-31B-it-Q4_K_M.gguf",
   "model_size": null,
   "quant": null,
   "runtime": "llama.cpp",
   "runtime_version": null,
   "workload": "chat-short",
   "suite_version": "suite-v1",
   "decode_tps": 67.37,
   "prefill_tps": null,
   "ttft_ms": 157.4,
   "prompt_tokens": 0,
   "output_tokens": 253,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_mln72x5zbis",
   "run_url": "https://llm-speed.com/r/r_mln72x5zbis",
   "date": "2026-07-01",
   "hardware": "RTX 5090 (32GB) + AMD Ryzen 7 9850X3D 8-Core Processor (8c) + 30GB",
   "model": "Qwen2.5-Coder-7B-Instruct",
   "model_size": null,
   "quant": null,
   "runtime": "llama.cpp",
   "runtime_version": null,
   "workload": "chat-short",
   "suite_version": "suite-v1",
   "decode_tps": 244.91,
   "prefill_tps": null,
   "ttft_ms": 36.1,
   "prompt_tokens": 0,
   "output_tokens": 256,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_mj85rtghwmg",
   "run_url": "https://llm-speed.com/r/r_mj85rtghwmg",
   "date": "2026-07-01",
   "hardware": "RTX 5090 (32GB) + AMD Ryzen 7 9850X3D 8-Core Processor (8c) + 30GB",
   "model": "Qwen3.6-27B-Q4_K_M.gguf",
   "model_size": null,
   "quant": null,
   "runtime": "llama.cpp",
   "runtime_version": null,
   "workload": "chat-short",
   "suite_version": "suite-v1",
   "decode_tps": 71.09,
   "prefill_tps": null,
   "ttft_ms": 241.0,
   "prompt_tokens": 0,
   "output_tokens": 256,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_lzv-y6_-n0b",
   "run_url": "https://llm-speed.com/r/r_lzv-y6_-n0b",
   "date": "2026-07-01",
   "hardware": "RTX 5090 (32GB) + AMD Ryzen 7 9850X3D 8-Core Processor (8c) + 30GB",
   "model": "gemma-4-31B-it-Q4_K_M.gguf",
   "model_size": null,
   "quant": null,
   "runtime": "llama.cpp",
   "runtime_version": null,
   "workload": "chat-short",
   "suite_version": "suite-v1",
   "decode_tps": 67.49,
   "prefill_tps": null,
   "ttft_ms": 220.7,
   "prompt_tokens": 0,
   "output_tokens": 253,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_lrxodemb-w0",
   "run_url": "https://llm-speed.com/r/r_lrxodemb-w0",
   "date": "2026-07-01",
   "hardware": "RTX 5090 (32GB) + AMD Ryzen 7 9850X3D 8-Core Processor (8c) + 30GB",
   "model": "gemma-4-31B-it-Q4_K_M.gguf",
   "model_size": null,
   "quant": null,
   "runtime": "llama.cpp",
   "runtime_version": null,
   "workload": "chat-short",
   "suite_version": "suite-v1",
   "decode_tps": 67.67,
   "prefill_tps": null,
   "ttft_ms": 208.6,
   "prompt_tokens": 0,
   "output_tokens": 253,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_lblt3fsiqzb",
   "run_url": "https://llm-speed.com/r/r_lblt3fsiqzb",
   "date": "2026-07-01",
   "hardware": "RTX 5090 (32GB) + AMD Ryzen 7 9850X3D 8-Core Processor (8c) + 30GB",
   "model": "Qwen3.6-27B-Q4_K_M.gguf",
   "model_size": null,
   "quant": null,
   "runtime": "llama.cpp",
   "runtime_version": null,
   "workload": "chat-short",
   "suite_version": "suite-v1",
   "decode_tps": 69.45,
   "prefill_tps": null,
   "ttft_ms": 174.8,
   "prompt_tokens": 0,
   "output_tokens": 256,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_l-wg5a6o-vg",
   "run_url": "https://llm-speed.com/r/r_l-wg5a6o-vg",
   "date": "2026-07-01",
   "hardware": "RTX 5090 (32GB) + AMD Ryzen 7 9850X3D 8-Core Processor (8c) + 30GB",
   "model": "Qwen3.6-27B-Q4_K_M.gguf",
   "model_size": null,
   "quant": null,
   "runtime": "llama.cpp",
   "runtime_version": null,
   "workload": "chat-short",
   "suite_version": "suite-v1",
   "decode_tps": 72.57,
   "prefill_tps": null,
   "ttft_ms": 171.8,
   "prompt_tokens": 0,
   "output_tokens": 256,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_kfrkg-vn376",
   "run_url": "https://llm-speed.com/r/r_kfrkg-vn376",
   "date": "2026-07-01",
   "hardware": "RTX 5090 (32GB) + AMD Ryzen 7 9850X3D 8-Core Processor (8c) + 30GB",
   "model": "Llama-3.1-8B-Instruct",
   "model_size": null,
   "quant": null,
   "runtime": "llama.cpp",
   "runtime_version": null,
   "workload": "chat-short",
   "suite_version": "suite-v1",
   "decode_tps": 232.22,
   "prefill_tps": null,
   "ttft_ms": 33.8,
   "prompt_tokens": 0,
   "output_tokens": 256,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_k1u_k1j_1i2",
   "run_url": "https://llm-speed.com/r/r_k1u_k1j_1i2",
   "date": "2026-07-01",
   "hardware": "RTX 5090 (32GB) + AMD Ryzen 7 9850X3D 8-Core Processor (8c) + 30GB",
   "model": "phi-4",
   "model_size": null,
   "quant": null,
   "runtime": "llama.cpp",
   "runtime_version": null,
   "workload": "concurrent-decode",
   "suite_version": "suite-v1",
   "decode_tps": 140.2,
   "prefill_tps": null,
   "ttft_ms": null,
   "prompt_tokens": 1024,
   "output_tokens": 256,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_k1u_k1j_1i2",
   "run_url": "https://llm-speed.com/r/r_k1u_k1j_1i2",
   "date": "2026-07-01",
   "hardware": "RTX 5090 (32GB) + AMD Ryzen 7 9850X3D 8-Core Processor (8c) + 30GB",
   "model": "phi-4",
   "model_size": null,
   "quant": null,
   "runtime": "llama.cpp",
   "runtime_version": null,
   "workload": "chat-short",
   "suite_version": "suite-v1",
   "decode_tps": 139.06,
   "prefill_tps": null,
   "ttft_ms": 51.1,
   "prompt_tokens": 0,
   "output_tokens": 256,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_k1u_k1j_1i2",
   "run_url": "https://llm-speed.com/r/r_k1u_k1j_1i2",
   "date": "2026-07-01",
   "hardware": "RTX 5090 (32GB) + AMD Ryzen 7 9850X3D 8-Core Processor (8c) + 30GB",
   "model": "phi-4",
   "model_size": null,
   "quant": null,
   "runtime": "llama.cpp",
   "runtime_version": null,
   "workload": "chat-long",
   "suite_version": "suite-v1",
   "decode_tps": 133.73,
   "prefill_tps": null,
   "ttft_ms": 432.8,
   "prompt_tokens": 0,
   "output_tokens": 410,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_k1u_k1j_1i2",
   "run_url": "https://llm-speed.com/r/r_k1u_k1j_1i2",
   "date": "2026-07-01",
   "hardware": "RTX 5090 (32GB) + AMD Ryzen 7 9850X3D 8-Core Processor (8c) + 30GB",
   "model": "phi-4",
   "model_size": null,
   "quant": null,
   "runtime": "llama.cpp",
   "runtime_version": null,
   "workload": "agent-trace",
   "suite_version": "suite-v1",
   "decode_tps": 124.8,
   "prefill_tps": 21162.44,
   "ttft_ms": 102.5,
   "prompt_tokens": 3362,
   "output_tokens": 840,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_k070lz99uzi",
   "run_url": "https://llm-speed.com/r/r_k070lz99uzi",
   "date": "2026-07-01",
   "hardware": "RTX 5090 (32GB) + AMD Ryzen 7 9850X3D 8-Core Processor (8c) + 30GB",
   "model": "Qwen3.6-35B-A3B-Q4_K_M.gguf",
   "model_size": null,
   "quant": null,
   "runtime": "llama.cpp",
   "runtime_version": null,
   "workload": "chat-short",
   "suite_version": "suite-v1",
   "decode_tps": 216.55,
   "prefill_tps": null,
   "ttft_ms": 184.7,
   "prompt_tokens": 0,
   "output_tokens": 256,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_jrwk2oj1el-",
   "run_url": "https://llm-speed.com/r/r_jrwk2oj1el-",
   "date": "2026-07-01",
   "hardware": "RTX 5090 (32GB) + AMD Ryzen 7 9850X3D 8-Core Processor (8c) + 30GB",
   "model": "Qwen3.6-27B-Q4_K_M.gguf",
   "model_size": null,
   "quant": null,
   "runtime": "llama.cpp",
   "runtime_version": null,
   "workload": "chat-short",
   "suite_version": "suite-v1",
   "decode_tps": 69.9,
   "prefill_tps": null,
   "ttft_ms": 180.4,
   "prompt_tokens": 0,
   "output_tokens": 256,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_jkn4ltg393p",
   "run_url": "https://llm-speed.com/r/r_jkn4ltg393p",
   "date": "2026-07-01",
   "hardware": "RTX 5090 (32GB) + AMD Ryzen 7 9850X3D 8-Core Processor (8c) + 30GB",
   "model": "gemma-4-31B-it-Q4_K_M.gguf",
   "model_size": null,
   "quant": null,
   "runtime": "llama.cpp",
   "runtime_version": null,
   "workload": "chat-short",
   "suite_version": "suite-v1",
   "decode_tps": 67.3,
   "prefill_tps": null,
   "ttft_ms": 167.1,
   "prompt_tokens": 0,
   "output_tokens": 253,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_jbit-ru039h",
   "run_url": "https://llm-speed.com/r/r_jbit-ru039h",
   "date": "2026-07-01",
   "hardware": "RTX 5090 (32GB) + AMD Ryzen 7 9850X3D 8-Core Processor (8c) + 30GB",
   "model": "Qwen3.6-27B-Q4_K_M.gguf",
   "model_size": null,
   "quant": null,
   "runtime": "llama.cpp",
   "runtime_version": null,
   "workload": "chat-short",
   "suite_version": "suite-v1",
   "decode_tps": 72.44,
   "prefill_tps": null,
   "ttft_ms": 172.4,
   "prompt_tokens": 0,
   "output_tokens": 256,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_j1t3p2tae3a",
   "run_url": "https://llm-speed.com/r/r_j1t3p2tae3a",
   "date": "2026-07-01",
   "hardware": "RTX 5090 (32GB) + AMD Ryzen 7 9850X3D 8-Core Processor (8c) + 30GB",
   "model": "Qwen3.6-27B-Q4_K_M.gguf",
   "model_size": null,
   "quant": null,
   "runtime": "llama.cpp",
   "runtime_version": null,
   "workload": "chat-short",
   "suite_version": "suite-v1",
   "decode_tps": 72.18,
   "prefill_tps": null,
   "ttft_ms": 288.9,
   "prompt_tokens": 0,
   "output_tokens": 256,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_j1cr6asrnn6",
   "run_url": "https://llm-speed.com/r/r_j1cr6asrnn6",
   "date": "2026-07-01",
   "hardware": "RTX 5090 (32GB) + AMD Ryzen 7 9850X3D 8-Core Processor (8c) + 30GB",
   "model": "Qwen3.6-27B-Q4_K_M.gguf",
   "model_size": null,
   "quant": null,
   "runtime": "llama.cpp",
   "runtime_version": null,
   "workload": "chat-short",
   "suite_version": "suite-v1",
   "decode_tps": 73.76,
   "prefill_tps": null,
   "ttft_ms": 346.4,
   "prompt_tokens": 0,
   "output_tokens": 256,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_j0xki0x2kn2",
   "run_url": "https://llm-speed.com/r/r_j0xki0x2kn2",
   "date": "2026-07-01",
   "hardware": "RTX 5090 (32GB) + AMD Ryzen 7 9850X3D 8-Core Processor (8c) + 30GB",
   "model": "Qwen3.6-27B-Q4_K_M.gguf",
   "model_size": null,
   "quant": null,
   "runtime": "llama.cpp",
   "runtime_version": null,
   "workload": "chat-short",
   "suite_version": "suite-v1",
   "decode_tps": 72.63,
   "prefill_tps": null,
   "ttft_ms": 169.6,
   "prompt_tokens": 0,
   "output_tokens": 256,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_hj-6scdybyq",
   "run_url": "https://llm-speed.com/r/r_hj-6scdybyq",
   "date": "2026-07-01",
   "hardware": "RTX 5090 (32GB) + AMD Ryzen 7 9850X3D 8-Core Processor (8c) + 30GB",
   "model": "Qwen3.6-35B-A3B-Q4_K_M.gguf",
   "model_size": null,
   "quant": null,
   "runtime": "llama.cpp",
   "runtime_version": null,
   "workload": "chat-short",
   "suite_version": "suite-v1",
   "decode_tps": 222.38,
   "prefill_tps": null,
   "ttft_ms": 145.0,
   "prompt_tokens": 0,
   "output_tokens": 256,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_gipbd9a8wrn",
   "run_url": "https://llm-speed.com/r/r_gipbd9a8wrn",
   "date": "2026-07-01",
   "hardware": "RTX 5090 (32GB) + AMD Ryzen 7 9850X3D 8-Core Processor (8c) + 30GB",
   "model": "Qwen3.6-27B-Q4_K_M.gguf",
   "model_size": null,
   "quant": null,
   "runtime": "llama.cpp",
   "runtime_version": null,
   "workload": "chat-short",
   "suite_version": "suite-v1",
   "decode_tps": 71.5,
   "prefill_tps": null,
   "ttft_ms": 212.8,
   "prompt_tokens": 0,
   "output_tokens": 256,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_g37y03ugmpk",
   "run_url": "https://llm-speed.com/r/r_g37y03ugmpk",
   "date": "2026-07-01",
   "hardware": "RTX 5090 (32GB) + AMD Ryzen 7 9850X3D 8-Core Processor (8c) + 30GB",
   "model": "Qwen3.6-27B-Q4_K_M.gguf",
   "model_size": null,
   "quant": null,
   "runtime": "llama.cpp",
   "runtime_version": null,
   "workload": "chat-short",
   "suite_version": "suite-v1",
   "decode_tps": 72.15,
   "prefill_tps": null,
   "ttft_ms": 170.8,
   "prompt_tokens": 0,
   "output_tokens": 256,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_g1aw3mizd32",
   "run_url": "https://llm-speed.com/r/r_g1aw3mizd32",
   "date": "2026-07-01",
   "hardware": "RTX 5090 (32GB) + AMD Ryzen 7 9850X3D 8-Core Processor (8c) + 30GB",
   "model": "Qwen3.6-35B-A3B-Q4_K_M.gguf",
   "model_size": null,
   "quant": null,
   "runtime": "llama.cpp",
   "runtime_version": null,
   "workload": "chat-short",
   "suite_version": "suite-v1",
   "decode_tps": 223.98,
   "prefill_tps": null,
   "ttft_ms": 169.5,
   "prompt_tokens": 0,
   "output_tokens": 256,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_fo83sbnxax4",
   "run_url": "https://llm-speed.com/r/r_fo83sbnxax4",
   "date": "2026-07-01",
   "hardware": "RTX 5090 (32GB) + AMD Ryzen 7 9850X3D 8-Core Processor (8c) + 30GB",
   "model": "Qwen3.6-27B-Q4_K_M.gguf",
   "model_size": null,
   "quant": null,
   "runtime": "llama.cpp",
   "runtime_version": null,
   "workload": "chat-short",
   "suite_version": "suite-v1",
   "decode_tps": 71.99,
   "prefill_tps": null,
   "ttft_ms": 317.1,
   "prompt_tokens": 0,
   "output_tokens": 256,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_f7ulllqu4vk",
   "run_url": "https://llm-speed.com/r/r_f7ulllqu4vk",
   "date": "2026-07-01",
   "hardware": "RTX 5090 (32GB) + AMD Ryzen 7 9850X3D 8-Core Processor (8c) + 30GB",
   "model": "Qwen3.6-27B-Q4_K_M.gguf",
   "model_size": null,
   "quant": null,
   "runtime": "llama.cpp",
   "runtime_version": null,
   "workload": "chat-short",
   "suite_version": "suite-v1",
   "decode_tps": 73.96,
   "prefill_tps": null,
   "ttft_ms": 190.4,
   "prompt_tokens": 0,
   "output_tokens": 256,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_f1wlemwip05",
   "run_url": "https://llm-speed.com/r/r_f1wlemwip05",
   "date": "2026-07-01",
   "hardware": "RTX 5090 (32GB) + AMD Ryzen 7 9850X3D 8-Core Processor (8c) + 30GB",
   "model": "gemma-4-31B-it-Q4_K_M.gguf",
   "model_size": null,
   "quant": null,
   "runtime": "llama.cpp",
   "runtime_version": null,
   "workload": "chat-short",
   "suite_version": "suite-v1",
   "decode_tps": 67.17,
   "prefill_tps": null,
   "ttft_ms": 202.0,
   "prompt_tokens": 0,
   "output_tokens": 253,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_f14_uxceu0r",
   "run_url": "https://llm-speed.com/r/r_f14_uxceu0r",
   "date": "2026-07-01",
   "hardware": "RTX 5090 (32GB) + AMD Ryzen 7 9850X3D 8-Core Processor (8c) + 30GB",
   "model": "Qwen3.6-35B-A3B-Q4_K_M.gguf",
   "model_size": null,
   "quant": null,
   "runtime": "llama.cpp",
   "runtime_version": null,
   "workload": "chat-short",
   "suite_version": "suite-v1",
   "decode_tps": 218.97,
   "prefill_tps": null,
   "ttft_ms": 151.7,
   "prompt_tokens": 0,
   "output_tokens": 256,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_f-uojs7vckb",
   "run_url": "https://llm-speed.com/r/r_f-uojs7vckb",
   "date": "2026-07-01",
   "hardware": "RTX 5090 (32GB) + AMD Ryzen 7 9850X3D 8-Core Processor (8c) + 30GB",
   "model": "gemma-4-31B-it-Q4_K_M.gguf",
   "model_size": null,
   "quant": null,
   "runtime": "llama.cpp",
   "runtime_version": null,
   "workload": "chat-short",
   "suite_version": "suite-v1",
   "decode_tps": 66.0,
   "prefill_tps": null,
   "ttft_ms": 226.2,
   "prompt_tokens": 0,
   "output_tokens": 253,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_es8a88t0ez0",
   "run_url": "https://llm-speed.com/r/r_es8a88t0ez0",
   "date": "2026-07-01",
   "hardware": "RTX 5090 (32GB) + AMD Ryzen 7 9850X3D 8-Core Processor (8c) + 30GB",
   "model": "Qwen3.6-27B-Q4_K_M.gguf",
   "model_size": null,
   "quant": null,
   "runtime": "llama.cpp",
   "runtime_version": null,
   "workload": "chat-short",
   "suite_version": "suite-v1",
   "decode_tps": 70.07,
   "prefill_tps": null,
   "ttft_ms": 172.8,
   "prompt_tokens": 0,
   "output_tokens": 256,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_ercjdbdw2gi",
   "run_url": "https://llm-speed.com/r/r_ercjdbdw2gi",
   "date": "2026-07-01",
   "hardware": "RTX 5090 (32GB) + AMD Ryzen 7 9850X3D 8-Core Processor (8c) + 30GB",
   "model": "Qwen3.6-27B-Q4_K_M.gguf",
   "model_size": null,
   "quant": null,
   "runtime": "llama.cpp",
   "runtime_version": null,
   "workload": "chat-short",
   "suite_version": "suite-v1",
   "decode_tps": 72.32,
   "prefill_tps": null,
   "ttft_ms": 289.6,
   "prompt_tokens": 0,
   "output_tokens": 256,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_epzfs8k8ohh",
   "run_url": "https://llm-speed.com/r/r_epzfs8k8ohh",
   "date": "2026-07-01",
   "hardware": "RTX 5090 (32GB) + AMD Ryzen 7 9850X3D 8-Core Processor (8c) + 30GB",
   "model": "gemma-4-31B-it-Q4_K_M.gguf",
   "model_size": null,
   "quant": null,
   "runtime": "llama.cpp",
   "runtime_version": null,
   "workload": "chat-short",
   "suite_version": "suite-v1",
   "decode_tps": 68.77,
   "prefill_tps": null,
   "ttft_ms": 199.8,
   "prompt_tokens": 0,
   "output_tokens": 253,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_ea3vckhh-cn",
   "run_url": "https://llm-speed.com/r/r_ea3vckhh-cn",
   "date": "2026-07-01",
   "hardware": "RTX 5090 (32GB) + AMD Ryzen 7 9850X3D 8-Core Processor (8c) + 30GB",
   "model": "gemma-4-31B-it-Q4_K_M.gguf",
   "model_size": null,
   "quant": null,
   "runtime": "llama.cpp",
   "runtime_version": null,
   "workload": "chat-short",
   "suite_version": "suite-v1",
   "decode_tps": 67.73,
   "prefill_tps": null,
   "ttft_ms": 227.8,
   "prompt_tokens": 0,
   "output_tokens": 253,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_ea1db-lhv0r",
   "run_url": "https://llm-speed.com/r/r_ea1db-lhv0r",
   "date": "2026-07-01",
   "hardware": "RTX 5090 (32GB) + AMD Ryzen 7 9850X3D 8-Core Processor (8c) + 30GB",
   "model": "Qwen3.6-27B-Q4_K_M.gguf",
   "model_size": null,
   "quant": null,
   "runtime": "llama.cpp",
   "runtime_version": null,
   "workload": "chat-short",
   "suite_version": "suite-v1",
   "decode_tps": 70.89,
   "prefill_tps": null,
   "ttft_ms": 185.4,
   "prompt_tokens": 0,
   "output_tokens": 256,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_e7_858msfny",
   "run_url": "https://llm-speed.com/r/r_e7_858msfny",
   "date": "2026-07-01",
   "hardware": "RTX 5090 (32GB) + AMD Ryzen 7 9850X3D 8-Core Processor (8c) + 30GB",
   "model": "Qwen3.6-27B-Q4_K_M.gguf",
   "model_size": null,
   "quant": null,
   "runtime": "llama.cpp",
   "runtime_version": null,
   "workload": "chat-short",
   "suite_version": "suite-v1",
   "decode_tps": 69.61,
   "prefill_tps": null,
   "ttft_ms": 231.0,
   "prompt_tokens": 0,
   "output_tokens": 256,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_e-k4aea8ipr",
   "run_url": "https://llm-speed.com/r/r_e-k4aea8ipr",
   "date": "2026-07-01",
   "hardware": "RTX 5090 (32GB) + AMD Ryzen 7 9850X3D 8-Core Processor (8c) + 30GB",
   "model": "phi-4",
   "model_size": null,
   "quant": null,
   "runtime": "llama.cpp",
   "runtime_version": null,
   "workload": "chat-short",
   "suite_version": "suite-v1",
   "decode_tps": 140.91,
   "prefill_tps": null,
   "ttft_ms": 56.7,
   "prompt_tokens": 0,
   "output_tokens": 256,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_dqelw_a8f2c",
   "run_url": "https://llm-speed.com/r/r_dqelw_a8f2c",
   "date": "2026-07-01",
   "hardware": "RTX 5090 (32GB) + AMD Ryzen 7 9850X3D 8-Core Processor (8c) + 30GB",
   "model": "Qwen3.6-35B-A3B-Q4_K_M.gguf",
   "model_size": null,
   "quant": null,
   "runtime": "llama.cpp",
   "runtime_version": null,
   "workload": "chat-short",
   "suite_version": "suite-v1",
   "decode_tps": 221.63,
   "prefill_tps": null,
   "ttft_ms": 207.1,
   "prompt_tokens": 0,
   "output_tokens": 256,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_dp6sr2_iwcf",
   "run_url": "https://llm-speed.com/r/r_dp6sr2_iwcf",
   "date": "2026-07-01",
   "hardware": "RTX 5090 (32GB) + AMD Ryzen 7 9850X3D 8-Core Processor (8c) + 30GB",
   "model": "Qwen3.6-35B-A3B-Q4_K_M.gguf",
   "model_size": null,
   "quant": null,
   "runtime": "llama.cpp",
   "runtime_version": null,
   "workload": "chat-short",
   "suite_version": "suite-v1",
   "decode_tps": 223.28,
   "prefill_tps": null,
   "ttft_ms": 263.2,
   "prompt_tokens": 0,
   "output_tokens": 256,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_dee_mcm1ga0",
   "run_url": "https://llm-speed.com/r/r_dee_mcm1ga0",
   "date": "2026-07-01",
   "hardware": "RTX 5090 (32GB) + AMD Ryzen 7 9850X3D 8-Core Processor (8c) + 30GB",
   "model": "gemma-4-31B-it-Q4_K_M.gguf",
   "model_size": null,
   "quant": null,
   "runtime": "llama.cpp",
   "runtime_version": null,
   "workload": "chat-short",
   "suite_version": "suite-v1",
   "decode_tps": 67.8,
   "prefill_tps": null,
   "ttft_ms": 189.4,
   "prompt_tokens": 0,
   "output_tokens": 253,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_d1ydh-i8f9j",
   "run_url": "https://llm-speed.com/r/r_d1ydh-i8f9j",
   "date": "2026-07-01",
   "hardware": "RTX 5090 (32GB) + AMD Ryzen 7 9850X3D 8-Core Processor (8c) + 30GB",
   "model": "Qwen3.6-35B-A3B-Q4_K_M.gguf",
   "model_size": null,
   "quant": null,
   "runtime": "llama.cpp",
   "runtime_version": null,
   "workload": "chat-short",
   "suite_version": "suite-v1",
   "decode_tps": 208.57,
   "prefill_tps": null,
   "ttft_ms": 219.9,
   "prompt_tokens": 0,
   "output_tokens": 256,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_cxuvzpfj-yn",
   "run_url": "https://llm-speed.com/r/r_cxuvzpfj-yn",
   "date": "2026-07-01",
   "hardware": "RTX 5090 (32GB) + AMD Ryzen 7 9850X3D 8-Core Processor (8c) + 30GB",
   "model": "Qwen3.6-35B-A3B-Q4_K_M.gguf",
   "model_size": null,
   "quant": null,
   "runtime": "llama.cpp",
   "runtime_version": null,
   "workload": "chat-short",
   "suite_version": "suite-v1",
   "decode_tps": 223.31,
   "prefill_tps": null,
   "ttft_ms": 153.7,
   "prompt_tokens": 0,
   "output_tokens": 256,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_carzhcoe-mi",
   "run_url": "https://llm-speed.com/r/r_carzhcoe-mi",
   "date": "2026-07-01",
   "hardware": "RTX 5090 (32GB) + AMD Ryzen 7 9850X3D 8-Core Processor (8c) + 30GB",
   "model": "Qwen3.6-27B-Q4_K_M.gguf",
   "model_size": null,
   "quant": null,
   "runtime": "llama.cpp",
   "runtime_version": null,
   "workload": "chat-short",
   "suite_version": "suite-v1",
   "decode_tps": 70.98,
   "prefill_tps": null,
   "ttft_ms": 199.6,
   "prompt_tokens": 0,
   "output_tokens": 256,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_c7qyvvmmsv1",
   "run_url": "https://llm-speed.com/r/r_c7qyvvmmsv1",
   "date": "2026-07-01",
   "hardware": "RTX 5090 (32GB) + AMD Ryzen 7 9850X3D 8-Core Processor (8c) + 30GB",
   "model": "Qwen3-Coder-30B-A3B-Instruct",
   "model_size": null,
   "quant": null,
   "runtime": "llama.cpp",
   "runtime_version": null,
   "workload": "chat-short",
   "suite_version": "suite-v1",
   "decode_tps": 259.91,
   "prefill_tps": null,
   "ttft_ms": 218.4,
   "prompt_tokens": 0,
   "output_tokens": 256,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_c6t0q3m_tti",
   "run_url": "https://llm-speed.com/r/r_c6t0q3m_tti",
   "date": "2026-07-01",
   "hardware": "RTX 5090 (32GB) + AMD Ryzen 7 9850X3D 8-Core Processor (8c) + 30GB",
   "model": "Qwen3.6-35B-A3B-Q4_K_M.gguf",
   "model_size": null,
   "quant": null,
   "runtime": "llama.cpp",
   "runtime_version": null,
   "workload": "chat-short",
   "suite_version": "suite-v1",
   "decode_tps": 216.24,
   "prefill_tps": null,
   "ttft_ms": 242.7,
   "prompt_tokens": 0,
   "output_tokens": 256,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_c25tn-sl4bn",
   "run_url": "https://llm-speed.com/r/r_c25tn-sl4bn",
   "date": "2026-07-01",
   "hardware": "RTX 5090 (32GB) + AMD Ryzen 7 9850X3D 8-Core Processor (8c) + 30GB",
   "model": "Qwen3.6-35B-A3B-Q4_K_M.gguf",
   "model_size": null,
   "quant": null,
   "runtime": "llama.cpp",
   "runtime_version": null,
   "workload": "chat-short",
   "suite_version": "suite-v1",
   "decode_tps": 221.77,
   "prefill_tps": null,
   "ttft_ms": 203.8,
   "prompt_tokens": 0,
   "output_tokens": 256,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_bjy5a5izxjc",
   "run_url": "https://llm-speed.com/r/r_bjy5a5izxjc",
   "date": "2026-07-01",
   "hardware": "RTX 5090 (32GB) + AMD Ryzen 7 9850X3D 8-Core Processor (8c) + 30GB",
   "model": "Qwen2.5-32B-Instruct",
   "model_size": null,
   "quant": null,
   "runtime": "llama.cpp",
   "runtime_version": null,
   "workload": "concurrent-decode",
   "suite_version": "suite-v1",
   "decode_tps": 68.6,
   "prefill_tps": null,
   "ttft_ms": null,
   "prompt_tokens": 1024,
   "output_tokens": 256,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_bjy5a5izxjc",
   "run_url": "https://llm-speed.com/r/r_bjy5a5izxjc",
   "date": "2026-07-01",
   "hardware": "RTX 5090 (32GB) + AMD Ryzen 7 9850X3D 8-Core Processor (8c) + 30GB",
   "model": "Qwen2.5-32B-Instruct",
   "model_size": null,
   "quant": null,
   "runtime": "llama.cpp",
   "runtime_version": null,
   "workload": "chat-short",
   "suite_version": "suite-v1",
   "decode_tps": 68.71,
   "prefill_tps": null,
   "ttft_ms": 77.0,
   "prompt_tokens": 0,
   "output_tokens": 256,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_bjy5a5izxjc",
   "run_url": "https://llm-speed.com/r/r_bjy5a5izxjc",
   "date": "2026-07-01",
   "hardware": "RTX 5090 (32GB) + AMD Ryzen 7 9850X3D 8-Core Processor (8c) + 30GB",
   "model": "Qwen2.5-32B-Instruct",
   "model_size": null,
   "quant": null,
   "runtime": "llama.cpp",
   "runtime_version": null,
   "workload": "chat-long",
   "suite_version": "suite-v1",
   "decode_tps": 65.36,
   "prefill_tps": null,
   "ttft_ms": 967.7,
   "prompt_tokens": 0,
   "output_tokens": 326,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_bjy5a5izxjc",
   "run_url": "https://llm-speed.com/r/r_bjy5a5izxjc",
   "date": "2026-07-01",
   "hardware": "RTX 5090 (32GB) + AMD Ryzen 7 9850X3D 8-Core Processor (8c) + 30GB",
   "model": "Qwen2.5-32B-Instruct",
   "model_size": null,
   "quant": null,
   "runtime": "llama.cpp",
   "runtime_version": null,
   "workload": "agent-trace",
   "suite_version": "suite-v1",
   "decode_tps": 66.45,
   "prefill_tps": 10407.9,
   "ttft_ms": 214.2,
   "prompt_tokens": 3362,
   "output_tokens": 840,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_bgylub2qqr-",
   "run_url": "https://llm-speed.com/r/r_bgylub2qqr-",
   "date": "2026-07-01",
   "hardware": "RTX 5090 (32GB) + AMD Ryzen 7 9850X3D 8-Core Processor (8c) + 30GB",
   "model": "Qwen3.6-27B-Q4_K_M.gguf",
   "model_size": null,
   "quant": null,
   "runtime": "llama.cpp",
   "runtime_version": null,
   "workload": "chat-short",
   "suite_version": "suite-v1",
   "decode_tps": 69.84,
   "prefill_tps": null,
   "ttft_ms": 398.5,
   "prompt_tokens": 0,
   "output_tokens": 256,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_bfpto9so2o1",
   "run_url": "https://llm-speed.com/r/r_bfpto9so2o1",
   "date": "2026-07-01",
   "hardware": "RTX 5090 (32GB) + AMD Ryzen 7 9850X3D 8-Core Processor (8c) + 30GB",
   "model": "DeepSeek-Coder-V2-Lite-Instruct",
   "model_size": null,
   "quant": null,
   "runtime": "llama.cpp",
   "runtime_version": null,
   "workload": "concurrent-decode",
   "suite_version": "suite-v1",
   "decode_tps": 268.41,
   "prefill_tps": null,
   "ttft_ms": null,
   "prompt_tokens": 1024,
   "output_tokens": 256,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_bfpto9so2o1",
   "run_url": "https://llm-speed.com/r/r_bfpto9so2o1",
   "date": "2026-07-01",
   "hardware": "RTX 5090 (32GB) + AMD Ryzen 7 9850X3D 8-Core Processor (8c) + 30GB",
   "model": "DeepSeek-Coder-V2-Lite-Instruct",
   "model_size": null,
   "quant": null,
   "runtime": "llama.cpp",
   "runtime_version": null,
   "workload": "chat-short",
   "suite_version": "suite-v1",
   "decode_tps": 293.14,
   "prefill_tps": null,
   "ttft_ms": 297.0,
   "prompt_tokens": 0,
   "output_tokens": 256,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_bfpto9so2o1",
   "run_url": "https://llm-speed.com/r/r_bfpto9so2o1",
   "date": "2026-07-01",
   "hardware": "RTX 5090 (32GB) + AMD Ryzen 7 9850X3D 8-Core Processor (8c) + 30GB",
   "model": "DeepSeek-Coder-V2-Lite-Instruct",
   "model_size": null,
   "quant": null,
   "runtime": "llama.cpp",
   "runtime_version": null,
   "workload": "chat-long",
   "suite_version": "suite-v1",
   "decode_tps": 181.17,
   "prefill_tps": null,
   "ttft_ms": 361.0,
   "prompt_tokens": 0,
   "output_tokens": 482,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_bfpto9so2o1",
   "run_url": "https://llm-speed.com/r/r_bfpto9so2o1",
   "date": "2026-07-01",
   "hardware": "RTX 5090 (32GB) + AMD Ryzen 7 9850X3D 8-Core Processor (8c) + 30GB",
   "model": "DeepSeek-Coder-V2-Lite-Instruct",
   "model_size": null,
   "quant": null,
   "runtime": "llama.cpp",
   "runtime_version": null,
   "workload": "agent-trace",
   "suite_version": "suite-v1",
   "decode_tps": 204.52,
   "prefill_tps": 24978.67,
   "ttft_ms": 80.7,
   "prompt_tokens": 3362,
   "output_tokens": 840,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_b9ul-vxh9sc",
   "run_url": "https://llm-speed.com/r/r_b9ul-vxh9sc",
   "date": "2026-07-01",
   "hardware": "RTX 5090 (32GB) + AMD Ryzen 7 9850X3D 8-Core Processor (8c) + 30GB",
   "model": "gpt-oss-20b",
   "model_size": null,
   "quant": null,
   "runtime": "llama.cpp",
   "runtime_version": null,
   "workload": "concurrent-decode",
   "suite_version": "suite-v1",
   "decode_tps": 302.77,
   "prefill_tps": null,
   "ttft_ms": null,
   "prompt_tokens": 1024,
   "output_tokens": 256,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_b9ul-vxh9sc",
   "run_url": "https://llm-speed.com/r/r_b9ul-vxh9sc",
   "date": "2026-07-01",
   "hardware": "RTX 5090 (32GB) + AMD Ryzen 7 9850X3D 8-Core Processor (8c) + 30GB",
   "model": "gpt-oss-20b",
   "model_size": null,
   "quant": null,
   "runtime": "llama.cpp",
   "runtime_version": null,
   "workload": "chat-short",
   "suite_version": "suite-v1",
   "decode_tps": 318.38,
   "prefill_tps": null,
   "ttft_ms": 172.5,
   "prompt_tokens": 0,
   "output_tokens": 253,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_b9ul-vxh9sc",
   "run_url": "https://llm-speed.com/r/r_b9ul-vxh9sc",
   "date": "2026-07-01",
   "hardware": "RTX 5090 (32GB) + AMD Ryzen 7 9850X3D 8-Core Processor (8c) + 30GB",
   "model": "gpt-oss-20b",
   "model_size": null,
   "quant": null,
   "runtime": "llama.cpp",
   "runtime_version": null,
   "workload": "chat-long",
   "suite_version": "suite-v1",
   "decode_tps": 301.62,
   "prefill_tps": null,
   "ttft_ms": 239.1,
   "prompt_tokens": 0,
   "output_tokens": 1015,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_b9ul-vxh9sc",
   "run_url": "https://llm-speed.com/r/r_b9ul-vxh9sc",
   "date": "2026-07-01",
   "hardware": "RTX 5090 (32GB) + AMD Ryzen 7 9850X3D 8-Core Processor (8c) + 30GB",
   "model": "gpt-oss-20b",
   "model_size": null,
   "quant": null,
   "runtime": "llama.cpp",
   "runtime_version": null,
   "workload": "agent-trace",
   "suite_version": "suite-v1",
   "decode_tps": 304.5,
   "prefill_tps": 26907.52,
   "ttft_ms": 77.7,
   "prompt_tokens": 3362,
   "output_tokens": 840,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_az687c4xasr",
   "run_url": "https://llm-speed.com/r/r_az687c4xasr",
   "date": "2026-07-01",
   "hardware": "RTX 5090 (32GB) + AMD Ryzen 7 9850X3D 8-Core Processor (8c) + 30GB",
   "model": "Qwen3.6-27B-Q4_K_M.gguf",
   "model_size": null,
   "quant": null,
   "runtime": "llama.cpp",
   "runtime_version": null,
   "workload": "chat-short",
   "suite_version": "suite-v1",
   "decode_tps": 73.56,
   "prefill_tps": null,
   "ttft_ms": 166.7,
   "prompt_tokens": 0,
   "output_tokens": 256,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_aj808r0dw53",
   "run_url": "https://llm-speed.com/r/r_aj808r0dw53",
   "date": "2026-07-01",
   "hardware": "RTX 5090 (32GB) + AMD Ryzen 7 9850X3D 8-Core Processor (8c) + 30GB",
   "model": "Qwen3.6-27B-Q4_K_M.gguf",
   "model_size": null,
   "quant": null,
   "runtime": "llama.cpp",
   "runtime_version": null,
   "workload": "chat-short",
   "suite_version": "suite-v1",
   "decode_tps": 69.6,
   "prefill_tps": null,
   "ttft_ms": 381.9,
   "prompt_tokens": 0,
   "output_tokens": 256,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_a9n_rervciy",
   "run_url": "https://llm-speed.com/r/r_a9n_rervciy",
   "date": "2026-07-01",
   "hardware": "RTX 5090 (32GB) + AMD Ryzen 7 9850X3D 8-Core Processor (8c) + 30GB",
   "model": "Qwen3.6-35B-A3B-Q4_K_M.gguf",
   "model_size": null,
   "quant": null,
   "runtime": "llama.cpp",
   "runtime_version": null,
   "workload": "chat-short",
   "suite_version": "suite-v1",
   "decode_tps": 212.8,
   "prefill_tps": null,
   "ttft_ms": 167.5,
   "prompt_tokens": 0,
   "output_tokens": 256,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_a56-wxl21lk",
   "run_url": "https://llm-speed.com/r/r_a56-wxl21lk",
   "date": "2026-07-01",
   "hardware": "RTX 5090 (32GB) + AMD Ryzen 7 9850X3D 8-Core Processor (8c) + 30GB",
   "model": "Qwen3.6-35B-A3B-Q4_K_M.gguf",
   "model_size": null,
   "quant": null,
   "runtime": "llama.cpp",
   "runtime_version": null,
   "workload": "chat-short",
   "suite_version": "suite-v1",
   "decode_tps": 223.97,
   "prefill_tps": null,
   "ttft_ms": 167.3,
   "prompt_tokens": 0,
   "output_tokens": 256,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r__b_bzmmab_8",
   "run_url": "https://llm-speed.com/r/r__b_bzmmab_8",
   "date": "2026-07-01",
   "hardware": "RTX 5090 (32GB) + AMD Ryzen 7 9850X3D 8-Core Processor (8c) + 30GB",
   "model": "gemma-2-9b-it",
   "model_size": null,
   "quant": null,
   "runtime": "llama.cpp",
   "runtime_version": null,
   "workload": "chat-short",
   "suite_version": "suite-v1",
   "decode_tps": 152.7,
   "prefill_tps": null,
   "ttft_ms": 58.9,
   "prompt_tokens": 0,
   "output_tokens": 256,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r__679a-4h44q",
   "run_url": "https://llm-speed.com/r/r__679a-4h44q",
   "date": "2026-07-01",
   "hardware": "RTX 5090 (32GB) + AMD Ryzen 7 9850X3D 8-Core Processor (8c) + 30GB",
   "model": "gemma-4-31B-it-Q4_K_M.gguf",
   "model_size": null,
   "quant": null,
   "runtime": "llama.cpp",
   "runtime_version": null,
   "workload": "chat-short",
   "suite_version": "suite-v1",
   "decode_tps": 66.7,
   "prefill_tps": null,
   "ttft_ms": 178.7,
   "prompt_tokens": 0,
   "output_tokens": 253,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_9ap2a34-le5",
   "run_url": "https://llm-speed.com/r/r_9ap2a34-le5",
   "date": "2026-07-01",
   "hardware": "RTX 5090 (32GB) + AMD Ryzen 7 9850X3D 8-Core Processor (8c) + 30GB",
   "model": "Qwen3.6-27B-Q4_K_M.gguf",
   "model_size": null,
   "quant": null,
   "runtime": "llama.cpp",
   "runtime_version": null,
   "workload": "chat-short",
   "suite_version": "suite-v1",
   "decode_tps": 69.84,
   "prefill_tps": null,
   "ttft_ms": 349.8,
   "prompt_tokens": 0,
   "output_tokens": 256,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_8qwzo79cf0s",
   "run_url": "https://llm-speed.com/r/r_8qwzo79cf0s",
   "date": "2026-07-01",
   "hardware": "RTX 5090 (32GB) + AMD Ryzen 7 9850X3D 8-Core Processor (8c) + 30GB",
   "model": "Qwen3.6-27B-Q4_K_M.gguf",
   "model_size": null,
   "quant": null,
   "runtime": "llama.cpp",
   "runtime_version": null,
   "workload": "chat-short",
   "suite_version": "suite-v1",
   "decode_tps": 72.59,
   "prefill_tps": null,
   "ttft_ms": 314.1,
   "prompt_tokens": 0,
   "output_tokens": 256,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_8l57cim1i10",
   "run_url": "https://llm-speed.com/r/r_8l57cim1i10",
   "date": "2026-07-01",
   "hardware": "RTX 5090 (32GB) + AMD Ryzen 7 9850X3D 8-Core Processor (8c) + 30GB",
   "model": "stable-code-instruct-3b",
   "model_size": null,
   "quant": null,
   "runtime": "llama.cpp",
   "runtime_version": null,
   "workload": "chat-short",
   "suite_version": "suite-v1",
   "decode_tps": 331.01,
   "prefill_tps": null,
   "ttft_ms": 36.7,
   "prompt_tokens": 0,
   "output_tokens": 219,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_83zbwf7c17p",
   "run_url": "https://llm-speed.com/r/r_83zbwf7c17p",
   "date": "2026-07-01",
   "hardware": "RTX 5090 (32GB) + AMD Ryzen 7 9850X3D 8-Core Processor (8c) + 30GB",
   "model": "Qwen3.6-27B-Q4_K_M.gguf",
   "model_size": null,
   "quant": null,
   "runtime": "llama.cpp",
   "runtime_version": null,
   "workload": "chat-short",
   "suite_version": "suite-v1",
   "decode_tps": 72.48,
   "prefill_tps": null,
   "ttft_ms": 287.8,
   "prompt_tokens": 0,
   "output_tokens": 256,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_7u98v6--uli",
   "run_url": "https://llm-speed.com/r/r_7u98v6--uli",
   "date": "2026-07-01",
   "hardware": "RTX 5090 (32GB) + AMD Ryzen 7 9850X3D 8-Core Processor (8c) + 30GB",
   "model": "Qwen3.6-27B-Q4_K_M.gguf",
   "model_size": null,
   "quant": null,
   "runtime": "llama.cpp",
   "runtime_version": null,
   "workload": "chat-short",
   "suite_version": "suite-v1",
   "decode_tps": 72.54,
   "prefill_tps": null,
   "ttft_ms": 170.9,
   "prompt_tokens": 0,
   "output_tokens": 256,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_7snxv0llk6f",
   "run_url": "https://llm-speed.com/r/r_7snxv0llk6f",
   "date": "2026-07-01",
   "hardware": "RTX 5090 (32GB) + AMD Ryzen 7 9850X3D 8-Core Processor (8c) + 30GB",
   "model": "Qwen3.6-27B-Q4_K_M.gguf",
   "model_size": null,
   "quant": null,
   "runtime": "llama.cpp",
   "runtime_version": null,
   "workload": "chat-short",
   "suite_version": "suite-v1",
   "decode_tps": 72.9,
   "prefill_tps": null,
   "ttft_ms": 164.8,
   "prompt_tokens": 0,
   "output_tokens": 256,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_616apthaana",
   "run_url": "https://llm-speed.com/r/r_616apthaana",
   "date": "2026-07-01",
   "hardware": "RTX 5090 (32GB) + AMD Ryzen 7 9850X3D 8-Core Processor (8c) + 30GB",
   "model": "gemma-4-31B-it-Q4_K_M.gguf",
   "model_size": null,
   "quant": null,
   "runtime": "llama.cpp",
   "runtime_version": null,
   "workload": "chat-short",
   "suite_version": "suite-v1",
   "decode_tps": 66.74,
   "prefill_tps": null,
   "ttft_ms": 199.7,
   "prompt_tokens": 0,
   "output_tokens": 253,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_608esz63ib8",
   "run_url": "https://llm-speed.com/r/r_608esz63ib8",
   "date": "2026-07-01",
   "hardware": "RTX 5090 (32GB) + AMD Ryzen 7 9850X3D 8-Core Processor (8c) + 30GB",
   "model": "gemma-4-31B-it-Q4_K_M.gguf",
   "model_size": null,
   "quant": null,
   "runtime": "llama.cpp",
   "runtime_version": null,
   "workload": "chat-short",
   "suite_version": "suite-v1",
   "decode_tps": 64.71,
   "prefill_tps": null,
   "ttft_ms": 197.3,
   "prompt_tokens": 0,
   "output_tokens": 253,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_5m7gbxjpw_c",
   "run_url": "https://llm-speed.com/r/r_5m7gbxjpw_c",
   "date": "2026-07-01",
   "hardware": "RTX 5090 (32GB) + AMD Ryzen 7 9850X3D 8-Core Processor (8c) + 30GB",
   "model": "Qwen3.6-27B-Q4_K_M.gguf",
   "model_size": null,
   "quant": null,
   "runtime": "llama.cpp",
   "runtime_version": null,
   "workload": "chat-short",
   "suite_version": "suite-v1",
   "decode_tps": 72.56,
   "prefill_tps": null,
   "ttft_ms": 215.0,
   "prompt_tokens": 0,
   "output_tokens": 256,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_5b-e5ortgcv",
   "run_url": "https://llm-speed.com/r/r_5b-e5ortgcv",
   "date": "2026-07-01",
   "hardware": "RTX 5090 (32GB) + AMD Ryzen 7 9850X3D 8-Core Processor (8c) + 30GB",
   "model": "Qwen3.6-27B-Q4_K_M.gguf",
   "model_size": null,
   "quant": null,
   "runtime": "llama.cpp",
   "runtime_version": null,
   "workload": "chat-short",
   "suite_version": "suite-v1",
   "decode_tps": 69.7,
   "prefill_tps": null,
   "ttft_ms": 166.6,
   "prompt_tokens": 0,
   "output_tokens": 256,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_5-8ewf89dyr",
   "run_url": "https://llm-speed.com/r/r_5-8ewf89dyr",
   "date": "2026-07-01",
   "hardware": "RTX 5090 (32GB) + AMD Ryzen 7 9850X3D 8-Core Processor (8c) + 30GB",
   "model": "Qwen3.6-27B-Q4_K_M.gguf",
   "model_size": null,
   "quant": null,
   "runtime": "llama.cpp",
   "runtime_version": null,
   "workload": "chat-short",
   "suite_version": "suite-v1",
   "decode_tps": 69.84,
   "prefill_tps": null,
   "ttft_ms": 342.7,
   "prompt_tokens": 0,
   "output_tokens": 256,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_4upad7ubcpi",
   "run_url": "https://llm-speed.com/r/r_4upad7ubcpi",
   "date": "2026-07-01",
   "hardware": "RTX 5090 (32GB) + AMD Ryzen 7 9850X3D 8-Core Processor (8c) + 30GB",
   "model": "Qwen3.6-27B-Q4_K_M.gguf",
   "model_size": null,
   "quant": null,
   "runtime": "llama.cpp",
   "runtime_version": null,
   "workload": "chat-short",
   "suite_version": "suite-v1",
   "decode_tps": 72.31,
   "prefill_tps": null,
   "ttft_ms": 198.3,
   "prompt_tokens": 0,
   "output_tokens": 256,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_4q040m4scic",
   "run_url": "https://llm-speed.com/r/r_4q040m4scic",
   "date": "2026-07-01",
   "hardware": "RTX 5090 (32GB) + AMD Ryzen 7 9850X3D 8-Core Processor (8c) + 30GB",
   "model": "Codestral-22B-v0.1",
   "model_size": null,
   "quant": null,
   "runtime": "llama.cpp",
   "runtime_version": null,
   "workload": "concurrent-decode",
   "suite_version": "suite-v1",
   "decode_tps": 100.27,
   "prefill_tps": null,
   "ttft_ms": null,
   "prompt_tokens": 1024,
   "output_tokens": 256,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_4q040m4scic",
   "run_url": "https://llm-speed.com/r/r_4q040m4scic",
   "date": "2026-07-01",
   "hardware": "RTX 5090 (32GB) + AMD Ryzen 7 9850X3D 8-Core Processor (8c) + 30GB",
   "model": "Codestral-22B-v0.1",
   "model_size": null,
   "quant": null,
   "runtime": "llama.cpp",
   "runtime_version": null,
   "workload": "chat-short",
   "suite_version": "suite-v1",
   "decode_tps": 93.72,
   "prefill_tps": null,
   "ttft_ms": 68.6,
   "prompt_tokens": 0,
   "output_tokens": 256,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_4q040m4scic",
   "run_url": "https://llm-speed.com/r/r_4q040m4scic",
   "date": "2026-07-01",
   "hardware": "RTX 5090 (32GB) + AMD Ryzen 7 9850X3D 8-Core Processor (8c) + 30GB",
   "model": "Codestral-22B-v0.1",
   "model_size": null,
   "quant": null,
   "runtime": "llama.cpp",
   "runtime_version": null,
   "workload": "chat-long",
   "suite_version": "suite-v1",
   "decode_tps": 87.36,
   "prefill_tps": null,
   "ttft_ms": 849.1,
   "prompt_tokens": 0,
   "output_tokens": 250,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_4q040m4scic",
   "run_url": "https://llm-speed.com/r/r_4q040m4scic",
   "date": "2026-07-01",
   "hardware": "RTX 5090 (32GB) + AMD Ryzen 7 9850X3D 8-Core Processor (8c) + 30GB",
   "model": "Codestral-22B-v0.1",
   "model_size": null,
   "quant": null,
   "runtime": "llama.cpp",
   "runtime_version": null,
   "workload": "agent-trace",
   "suite_version": "suite-v1",
   "decode_tps": 93.44,
   "prefill_tps": 10351.05,
   "ttft_ms": 190.8,
   "prompt_tokens": 3362,
   "output_tokens": 840,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_4puowxcjcfu",
   "run_url": "https://llm-speed.com/r/r_4puowxcjcfu",
   "date": "2026-07-01",
   "hardware": "RTX 5090 (32GB) + AMD Ryzen 7 9850X3D 8-Core Processor (8c) + 30GB",
   "model": "Qwen3.6-27B-Q4_K_M.gguf",
   "model_size": null,
   "quant": null,
   "runtime": "llama.cpp",
   "runtime_version": null,
   "workload": "chat-short",
   "suite_version": "suite-v1",
   "decode_tps": 69.34,
   "prefill_tps": null,
   "ttft_ms": 301.0,
   "prompt_tokens": 0,
   "output_tokens": 256,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_4fxpo_ja16m",
   "run_url": "https://llm-speed.com/r/r_4fxpo_ja16m",
   "date": "2026-07-01",
   "hardware": "RTX 5090 (32GB) + AMD Ryzen 7 9850X3D 8-Core Processor (8c) + 30GB",
   "model": "Qwen3.6-27B-Q4_K_M.gguf",
   "model_size": null,
   "quant": null,
   "runtime": "llama.cpp",
   "runtime_version": null,
   "workload": "chat-short",
   "suite_version": "suite-v1",
   "decode_tps": 69.05,
   "prefill_tps": null,
   "ttft_ms": 191.0,
   "prompt_tokens": 0,
   "output_tokens": 256,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_4fi6u0-u2ih",
   "run_url": "https://llm-speed.com/r/r_4fi6u0-u2ih",
   "date": "2026-07-01",
   "hardware": "RTX 5090 (32GB) + AMD Ryzen 7 9850X3D 8-Core Processor (8c) + 30GB",
   "model": "gemma-4-31B-it-Q4_K_M.gguf",
   "model_size": null,
   "quant": null,
   "runtime": "llama.cpp",
   "runtime_version": null,
   "workload": "chat-short",
   "suite_version": "suite-v1",
   "decode_tps": 67.24,
   "prefill_tps": null,
   "ttft_ms": 223.2,
   "prompt_tokens": 0,
   "output_tokens": 253,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_432qs45nio9",
   "run_url": "https://llm-speed.com/r/r_432qs45nio9",
   "date": "2026-07-01",
   "hardware": "RTX 5090 (32GB) + AMD Ryzen 7 9850X3D 8-Core Processor (8c) + 30GB",
   "model": "Qwen3.6-27B-Q4_K_M.gguf",
   "model_size": null,
   "quant": null,
   "runtime": "llama.cpp",
   "runtime_version": null,
   "workload": "chat-short",
   "suite_version": "suite-v1",
   "decode_tps": 72.27,
   "prefill_tps": null,
   "ttft_ms": 181.9,
   "prompt_tokens": 0,
   "output_tokens": 256,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_3yn-4321hp-",
   "run_url": "https://llm-speed.com/r/r_3yn-4321hp-",
   "date": "2026-07-01",
   "hardware": "RTX 5090 (32GB) + AMD Ryzen 7 9850X3D 8-Core Processor (8c) + 30GB",
   "model": "Qwen2.5-7B-Instruct",
   "model_size": null,
   "quant": null,
   "runtime": "llama.cpp",
   "runtime_version": null,
   "workload": "chat-short",
   "suite_version": "suite-v1",
   "decode_tps": 246.65,
   "prefill_tps": null,
   "ttft_ms": 33.9,
   "prompt_tokens": 0,
   "output_tokens": 256,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_3un_x00tm0t",
   "run_url": "https://llm-speed.com/r/r_3un_x00tm0t",
   "date": "2026-07-01",
   "hardware": "RTX 5090 (32GB) + AMD Ryzen 7 9850X3D 8-Core Processor (8c) + 30GB",
   "model": "Qwen3.6-27B-Q4_K_M.gguf",
   "model_size": null,
   "quant": null,
   "runtime": "llama.cpp",
   "runtime_version": null,
   "workload": "chat-short",
   "suite_version": "suite-v1",
   "decode_tps": 72.49,
   "prefill_tps": null,
   "ttft_ms": 178.2,
   "prompt_tokens": 0,
   "output_tokens": 256,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_3n3hnvdgwnx",
   "run_url": "https://llm-speed.com/r/r_3n3hnvdgwnx",
   "date": "2026-07-01",
   "hardware": "RTX 5090 (32GB) + AMD Ryzen 7 9850X3D 8-Core Processor (8c) + 30GB",
   "model": "Qwen3.6-27B-Q4_K_M.gguf",
   "model_size": null,
   "quant": null,
   "runtime": "llama.cpp",
   "runtime_version": null,
   "workload": "chat-short",
   "suite_version": "suite-v1",
   "decode_tps": 69.55,
   "prefill_tps": null,
   "ttft_ms": 179.5,
   "prompt_tokens": 0,
   "output_tokens": 256,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_3hda1iz7yd8",
   "run_url": "https://llm-speed.com/r/r_3hda1iz7yd8",
   "date": "2026-07-01",
   "hardware": "RTX 5090 (32GB) + AMD Ryzen 7 9850X3D 8-Core Processor (8c) + 30GB",
   "model": "Qwen3.6-27B-Q4_K_M.gguf",
   "model_size": null,
   "quant": null,
   "runtime": "llama.cpp",
   "runtime_version": null,
   "workload": "chat-short",
   "suite_version": "suite-v1",
   "decode_tps": 71.81,
   "prefill_tps": null,
   "ttft_ms": 176.7,
   "prompt_tokens": 0,
   "output_tokens": 256,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_36c-t9il61l",
   "run_url": "https://llm-speed.com/r/r_36c-t9il61l",
   "date": "2026-07-01",
   "hardware": "RTX 5090 (32GB) + AMD Ryzen 7 9850X3D 8-Core Processor (8c) + 30GB",
   "model": "Qwen3.6-35B-A3B-Q4_K_M.gguf",
   "model_size": null,
   "quant": null,
   "runtime": "llama.cpp",
   "runtime_version": null,
   "workload": "chat-short",
   "suite_version": "suite-v1",
   "decode_tps": 221.03,
   "prefill_tps": null,
   "ttft_ms": 165.5,
   "prompt_tokens": 0,
   "output_tokens": 256,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_2o9kb8hy9uk",
   "run_url": "https://llm-speed.com/r/r_2o9kb8hy9uk",
   "date": "2026-07-01",
   "hardware": "RTX 5090 (32GB) + AMD Ryzen 7 9850X3D 8-Core Processor (8c) + 30GB",
   "model": "Qwen3.6-27B-Q4_K_M.gguf",
   "model_size": null,
   "quant": null,
   "runtime": "llama.cpp",
   "runtime_version": null,
   "workload": "chat-short",
   "suite_version": "suite-v1",
   "decode_tps": 68.61,
   "prefill_tps": null,
   "ttft_ms": 183.3,
   "prompt_tokens": 0,
   "output_tokens": 256,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_2g-h0epeovp",
   "run_url": "https://llm-speed.com/r/r_2g-h0epeovp",
   "date": "2026-07-01",
   "hardware": "RTX 5090 (32GB) + AMD Ryzen 7 9850X3D 8-Core Processor (8c) + 30GB",
   "model": "Qwen3.6-27B-Q4_K_M.gguf",
   "model_size": null,
   "quant": null,
   "runtime": "llama.cpp",
   "runtime_version": null,
   "workload": "chat-short",
   "suite_version": "suite-v1",
   "decode_tps": 72.68,
   "prefill_tps": null,
   "ttft_ms": 165.9,
   "prompt_tokens": 0,
   "output_tokens": 256,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_2b9o6y_49mi",
   "run_url": "https://llm-speed.com/r/r_2b9o6y_49mi",
   "date": "2026-07-01",
   "hardware": "RTX 5090 (32GB) + AMD Ryzen 7 9850X3D 8-Core Processor (8c) + 30GB",
   "model": "Qwen2.5-Coder-7B-Instruct",
   "model_size": null,
   "quant": null,
   "runtime": "llama.cpp",
   "runtime_version": null,
   "workload": "concurrent-decode",
   "suite_version": "suite-v1",
   "decode_tps": 254.04,
   "prefill_tps": null,
   "ttft_ms": null,
   "prompt_tokens": 1024,
   "output_tokens": 256,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_2b9o6y_49mi",
   "run_url": "https://llm-speed.com/r/r_2b9o6y_49mi",
   "date": "2026-07-01",
   "hardware": "RTX 5090 (32GB) + AMD Ryzen 7 9850X3D 8-Core Processor (8c) + 30GB",
   "model": "Qwen2.5-Coder-7B-Instruct",
   "model_size": null,
   "quant": null,
   "runtime": "llama.cpp",
   "runtime_version": null,
   "workload": "chat-short",
   "suite_version": "suite-v1",
   "decode_tps": 255.3,
   "prefill_tps": null,
   "ttft_ms": 35.4,
   "prompt_tokens": 0,
   "output_tokens": 256,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_2b9o6y_49mi",
   "run_url": "https://llm-speed.com/r/r_2b9o6y_49mi",
   "date": "2026-07-01",
   "hardware": "RTX 5090 (32GB) + AMD Ryzen 7 9850X3D 8-Core Processor (8c) + 30GB",
   "model": "Qwen2.5-Coder-7B-Instruct",
   "model_size": null,
   "quant": null,
   "runtime": "llama.cpp",
   "runtime_version": null,
   "workload": "chat-long",
   "suite_version": "suite-v1",
   "decode_tps": 244.01,
   "prefill_tps": null,
   "ttft_ms": 228.0,
   "prompt_tokens": 0,
   "output_tokens": 489,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_2b9o6y_49mi",
   "run_url": "https://llm-speed.com/r/r_2b9o6y_49mi",
   "date": "2026-07-01",
   "hardware": "RTX 5090 (32GB) + AMD Ryzen 7 9850X3D 8-Core Processor (8c) + 30GB",
   "model": "Qwen2.5-Coder-7B-Instruct",
   "model_size": null,
   "quant": null,
   "runtime": "llama.cpp",
   "runtime_version": null,
   "workload": "agent-trace",
   "suite_version": "suite-v1",
   "decode_tps": 229.9,
   "prefill_tps": 38735.22,
   "ttft_ms": 57.4,
   "prompt_tokens": 3362,
   "output_tokens": 840,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_2673sv7x5m4",
   "run_url": "https://llm-speed.com/r/r_2673sv7x5m4",
   "date": "2026-07-01",
   "hardware": "RTX 5090 (32GB) + AMD Ryzen 7 9850X3D 8-Core Processor (8c) + 30GB",
   "model": "Qwen3.6-27B-Q4_K_M.gguf",
   "model_size": null,
   "quant": null,
   "runtime": "llama.cpp",
   "runtime_version": null,
   "workload": "chat-short",
   "suite_version": "suite-v1",
   "decode_tps": 69.68,
   "prefill_tps": null,
   "ttft_ms": 177.9,
   "prompt_tokens": 0,
   "output_tokens": 256,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_23fireoga9y",
   "run_url": "https://llm-speed.com/r/r_23fireoga9y",
   "date": "2026-07-01",
   "hardware": "RTX 5090 (32GB) + AMD Ryzen 7 9850X3D 8-Core Processor (8c) + 30GB",
   "model": "gemma-4-31B-it-Q4_K_M.gguf",
   "model_size": null,
   "quant": null,
   "runtime": "llama.cpp",
   "runtime_version": null,
   "workload": "chat-short",
   "suite_version": "suite-v1",
   "decode_tps": 67.27,
   "prefill_tps": null,
   "ttft_ms": 343.9,
   "prompt_tokens": 0,
   "output_tokens": 253,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_1shiviswt3d",
   "run_url": "https://llm-speed.com/r/r_1shiviswt3d",
   "date": "2026-07-01",
   "hardware": "RTX 5090 (32GB) + AMD Ryzen 7 9850X3D 8-Core Processor (8c) + 30GB",
   "model": "Qwen2.5-7B-Instruct",
   "model_size": null,
   "quant": null,
   "runtime": "llama.cpp",
   "runtime_version": null,
   "workload": "concurrent-decode",
   "suite_version": "suite-v1",
   "decode_tps": 240.67,
   "prefill_tps": null,
   "ttft_ms": null,
   "prompt_tokens": 1024,
   "output_tokens": 256,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_1shiviswt3d",
   "run_url": "https://llm-speed.com/r/r_1shiviswt3d",
   "date": "2026-07-01",
   "hardware": "RTX 5090 (32GB) + AMD Ryzen 7 9850X3D 8-Core Processor (8c) + 30GB",
   "model": "Qwen2.5-7B-Instruct",
   "model_size": null,
   "quant": null,
   "runtime": "llama.cpp",
   "runtime_version": null,
   "workload": "chat-short",
   "suite_version": "suite-v1",
   "decode_tps": 232.66,
   "prefill_tps": null,
   "ttft_ms": 37.0,
   "prompt_tokens": 0,
   "output_tokens": 256,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_1shiviswt3d",
   "run_url": "https://llm-speed.com/r/r_1shiviswt3d",
   "date": "2026-07-01",
   "hardware": "RTX 5090 (32GB) + AMD Ryzen 7 9850X3D 8-Core Processor (8c) + 30GB",
   "model": "Qwen2.5-7B-Instruct",
   "model_size": null,
   "quant": null,
   "runtime": "llama.cpp",
   "runtime_version": null,
   "workload": "chat-long",
   "suite_version": "suite-v1",
   "decode_tps": 232.4,
   "prefill_tps": null,
   "ttft_ms": 239.4,
   "prompt_tokens": 0,
   "output_tokens": 377,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_1shiviswt3d",
   "run_url": "https://llm-speed.com/r/r_1shiviswt3d",
   "date": "2026-07-01",
   "hardware": "RTX 5090 (32GB) + AMD Ryzen 7 9850X3D 8-Core Processor (8c) + 30GB",
   "model": "Qwen2.5-7B-Instruct",
   "model_size": null,
   "quant": null,
   "runtime": "llama.cpp",
   "runtime_version": null,
   "workload": "agent-trace",
   "suite_version": "suite-v1",
   "decode_tps": 232.53,
   "prefill_tps": 38007.68,
   "ttft_ms": 57.1,
   "prompt_tokens": 3362,
   "output_tokens": 840,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_1ofpvf4m7p5",
   "run_url": "https://llm-speed.com/r/r_1ofpvf4m7p5",
   "date": "2026-07-01",
   "hardware": "RTX 5090 (32GB) + AMD Ryzen 7 9850X3D 8-Core Processor (8c) + 30GB",
   "model": "gemma-4-31B-it-Q4_K_M.gguf",
   "model_size": null,
   "quant": null,
   "runtime": "llama.cpp",
   "runtime_version": null,
   "workload": "chat-short",
   "suite_version": "suite-v1",
   "decode_tps": 67.33,
   "prefill_tps": null,
   "ttft_ms": 155.9,
   "prompt_tokens": 0,
   "output_tokens": 253,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_1_xl4zb5-xj",
   "run_url": "https://llm-speed.com/r/r_1_xl4zb5-xj",
   "date": "2026-07-01",
   "hardware": "RTX 5090 (32GB) + AMD Ryzen 7 9850X3D 8-Core Processor (8c) + 30GB",
   "model": "gemma-2-9b-it",
   "model_size": null,
   "quant": null,
   "runtime": "llama.cpp",
   "runtime_version": null,
   "workload": "concurrent-decode",
   "suite_version": "suite-v1",
   "decode_tps": 68.57,
   "prefill_tps": null,
   "ttft_ms": null,
   "prompt_tokens": 1024,
   "output_tokens": 256,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_1_xl4zb5-xj",
   "run_url": "https://llm-speed.com/r/r_1_xl4zb5-xj",
   "date": "2026-07-01",
   "hardware": "RTX 5090 (32GB) + AMD Ryzen 7 9850X3D 8-Core Processor (8c) + 30GB",
   "model": "gemma-2-9b-it",
   "model_size": null,
   "quant": null,
   "runtime": "llama.cpp",
   "runtime_version": null,
   "workload": "chat-short",
   "suite_version": "suite-v1",
   "decode_tps": 69.45,
   "prefill_tps": null,
   "ttft_ms": 324.6,
   "prompt_tokens": 0,
   "output_tokens": 256,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_1_xl4zb5-xj",
   "run_url": "https://llm-speed.com/r/r_1_xl4zb5-xj",
   "date": "2026-07-01",
   "hardware": "RTX 5090 (32GB) + AMD Ryzen 7 9850X3D 8-Core Processor (8c) + 30GB",
   "model": "gemma-2-9b-it",
   "model_size": null,
   "quant": null,
   "runtime": "llama.cpp",
   "runtime_version": null,
   "workload": "chat-long",
   "suite_version": "suite-v1",
   "decode_tps": 69.42,
   "prefill_tps": null,
   "ttft_ms": 1044.2,
   "prompt_tokens": 0,
   "output_tokens": 1024,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_1_xl4zb5-xj",
   "run_url": "https://llm-speed.com/r/r_1_xl4zb5-xj",
   "date": "2026-07-01",
   "hardware": "RTX 5090 (32GB) + AMD Ryzen 7 9850X3D 8-Core Processor (8c) + 30GB",
   "model": "gemma-2-9b-it",
   "model_size": null,
   "quant": null,
   "runtime": "llama.cpp",
   "runtime_version": null,
   "workload": "agent-trace",
   "suite_version": "suite-v1",
   "decode_tps": 68.46,
   "prefill_tps": 4816.79,
   "ttft_ms": 420.2,
   "prompt_tokens": 3362,
   "output_tokens": 840,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_148vtu_7e08",
   "run_url": "https://llm-speed.com/r/r_148vtu_7e08",
   "date": "2026-07-01",
   "hardware": "RTX 5090 (32GB) + AMD Ryzen 7 9850X3D 8-Core Processor (8c) + 30GB",
   "model": "Qwen3.6-35B-A3B-Q4_K_M.gguf",
   "model_size": null,
   "quant": null,
   "runtime": "llama.cpp",
   "runtime_version": null,
   "workload": "chat-short",
   "suite_version": "suite-v1",
   "decode_tps": 217.36,
   "prefill_tps": null,
   "ttft_ms": 199.4,
   "prompt_tokens": 0,
   "output_tokens": 256,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_10yvku19-xt",
   "run_url": "https://llm-speed.com/r/r_10yvku19-xt",
   "date": "2026-07-01",
   "hardware": "RTX 5090 (32GB) + AMD Ryzen 7 9850X3D 8-Core Processor (8c) + 30GB",
   "model": "Qwen3.6-35B-A3B-Q4_K_M.gguf",
   "model_size": null,
   "quant": null,
   "runtime": "llama.cpp",
   "runtime_version": null,
   "workload": "chat-short",
   "suite_version": "suite-v1",
   "decode_tps": 223.81,
   "prefill_tps": null,
   "ttft_ms": 280.2,
   "prompt_tokens": 0,
   "output_tokens": 256,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_0ztzrxt5qw2",
   "run_url": "https://llm-speed.com/r/r_0ztzrxt5qw2",
   "date": "2026-07-01",
   "hardware": "RTX 5090 (32GB) + AMD Ryzen 7 9850X3D 8-Core Processor (8c) + 30GB",
   "model": "gemma-4-31B-it-Q4_K_M.gguf",
   "model_size": null,
   "quant": null,
   "runtime": "llama.cpp",
   "runtime_version": null,
   "workload": "chat-short",
   "suite_version": "suite-v1",
   "decode_tps": 67.97,
   "prefill_tps": null,
   "ttft_ms": 150.6,
   "prompt_tokens": 0,
   "output_tokens": 253,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_0igp5bsgx_7",
   "run_url": "https://llm-speed.com/r/r_0igp5bsgx_7",
   "date": "2026-07-01",
   "hardware": "RTX 5090 (32GB) + AMD Ryzen 7 9850X3D 8-Core Processor (8c) + 30GB",
   "model": "gemma-4-31B-it-Q4_K_M.gguf",
   "model_size": null,
   "quant": null,
   "runtime": "llama.cpp",
   "runtime_version": null,
   "workload": "chat-short",
   "suite_version": "suite-v1",
   "decode_tps": 67.44,
   "prefill_tps": null,
   "ttft_ms": 160.5,
   "prompt_tokens": 0,
   "output_tokens": 253,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_0dfevy9f4gh",
   "run_url": "https://llm-speed.com/r/r_0dfevy9f4gh",
   "date": "2026-07-01",
   "hardware": "RTX 5090 (32GB) + AMD Ryzen 7 9850X3D 8-Core Processor (8c) + 30GB",
   "model": "gemma-4-31B-it-Q4_K_M.gguf",
   "model_size": null,
   "quant": null,
   "runtime": "llama.cpp",
   "runtime_version": null,
   "workload": "chat-short",
   "suite_version": "suite-v1",
   "decode_tps": 67.05,
   "prefill_tps": null,
   "ttft_ms": 322.0,
   "prompt_tokens": 0,
   "output_tokens": 253,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_0_gs1rgl2fl",
   "run_url": "https://llm-speed.com/r/r_0_gs1rgl2fl",
   "date": "2026-07-01",
   "hardware": "RTX 5090 (32GB) + AMD Ryzen 7 9850X3D 8-Core Processor (8c) + 30GB",
   "model": "DeepSeek-Coder-V2-Lite-Instruct",
   "model_size": null,
   "quant": null,
   "runtime": "llama.cpp",
   "runtime_version": null,
   "workload": "chat-short",
   "suite_version": "suite-v1",
   "decode_tps": 309.48,
   "prefill_tps": null,
   "ttft_ms": 267.6,
   "prompt_tokens": 0,
   "output_tokens": 256,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_0-uc27u4l1s",
   "run_url": "https://llm-speed.com/r/r_0-uc27u4l1s",
   "date": "2026-07-01",
   "hardware": "RTX 5090 (32GB) + AMD Ryzen 7 9850X3D 8-Core Processor (8c) + 30GB",
   "model": "Qwen3.6-35B-A3B-Q4_K_M.gguf",
   "model_size": null,
   "quant": null,
   "runtime": "llama.cpp",
   "runtime_version": null,
   "workload": "chat-short",
   "suite_version": "suite-v1",
   "decode_tps": 217.54,
   "prefill_tps": null,
   "ttft_ms": 287.9,
   "prompt_tokens": 0,
   "output_tokens": 256,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_-txe_hiq44n",
   "run_url": "https://llm-speed.com/r/r_-txe_hiq44n",
   "date": "2026-07-01",
   "hardware": "RTX 5090 (32GB) + AMD Ryzen 7 9850X3D 8-Core Processor (8c) + 30GB",
   "model": "Qwen3-32B",
   "model_size": null,
   "quant": null,
   "runtime": "llama.cpp",
   "runtime_version": null,
   "workload": "concurrent-decode",
   "suite_version": "suite-v1",
   "decode_tps": 66.64,
   "prefill_tps": null,
   "ttft_ms": null,
   "prompt_tokens": 1024,
   "output_tokens": 256,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_-txe_hiq44n",
   "run_url": "https://llm-speed.com/r/r_-txe_hiq44n",
   "date": "2026-07-01",
   "hardware": "RTX 5090 (32GB) + AMD Ryzen 7 9850X3D 8-Core Processor (8c) + 30GB",
   "model": "Qwen3-32B",
   "model_size": null,
   "quant": null,
   "runtime": "llama.cpp",
   "runtime_version": null,
   "workload": "chat-short",
   "suite_version": "suite-v1",
   "decode_tps": 66.51,
   "prefill_tps": null,
   "ttft_ms": 158.0,
   "prompt_tokens": 0,
   "output_tokens": 254,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_-txe_hiq44n",
   "run_url": "https://llm-speed.com/r/r_-txe_hiq44n",
   "date": "2026-07-01",
   "hardware": "RTX 5090 (32GB) + AMD Ryzen 7 9850X3D 8-Core Processor (8c) + 30GB",
   "model": "Qwen3-32B",
   "model_size": null,
   "quant": null,
   "runtime": "llama.cpp",
   "runtime_version": null,
   "workload": "chat-long",
   "suite_version": "suite-v1",
   "decode_tps": 62.3,
   "prefill_tps": null,
   "ttft_ms": 1067.1,
   "prompt_tokens": 0,
   "output_tokens": 1020,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_-txe_hiq44n",
   "run_url": "https://llm-speed.com/r/r_-txe_hiq44n",
   "date": "2026-07-01",
   "hardware": "RTX 5090 (32GB) + AMD Ryzen 7 9850X3D 8-Core Processor (8c) + 30GB",
   "model": "Qwen3-32B",
   "model_size": null,
   "quant": null,
   "runtime": "llama.cpp",
   "runtime_version": null,
   "workload": "agent-trace",
   "suite_version": "suite-v1",
   "decode_tps": 64.34,
   "prefill_tps": 8394.07,
   "ttft_ms": 258.6,
   "prompt_tokens": 3362,
   "output_tokens": 840,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_-dxzsj2j-y_",
   "run_url": "https://llm-speed.com/r/r_-dxzsj2j-y_",
   "date": "2026-07-01",
   "hardware": "RTX 5090 (32GB) + AMD Ryzen 7 9850X3D 8-Core Processor (8c) + 30GB",
   "model": "gemma-4-31B-it-Q4_K_M.gguf",
   "model_size": null,
   "quant": null,
   "runtime": "llama.cpp",
   "runtime_version": null,
   "workload": "chat-short",
   "suite_version": "suite-v1",
   "decode_tps": 67.17,
   "prefill_tps": null,
   "ttft_ms": 202.2,
   "prompt_tokens": 0,
   "output_tokens": 253,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_-cpk_tabogc",
   "run_url": "https://llm-speed.com/r/r_-cpk_tabogc",
   "date": "2026-07-01",
   "hardware": "RTX 5090 (32GB) + AMD Ryzen 7 9850X3D 8-Core Processor (8c) + 30GB",
   "model": "Qwen3.6-27B-Q4_K_M.gguf",
   "model_size": null,
   "quant": null,
   "runtime": "llama.cpp",
   "runtime_version": null,
   "workload": "chat-short",
   "suite_version": "suite-v1",
   "decode_tps": 68.97,
   "prefill_tps": null,
   "ttft_ms": 188.6,
   "prompt_tokens": 0,
   "output_tokens": 256,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_-_ag61ig4f_",
   "run_url": "https://llm-speed.com/r/r_-_ag61ig4f_",
   "date": "2026-07-01",
   "hardware": "RTX 5090 (32GB) + AMD Ryzen 7 9850X3D 8-Core Processor (8c) + 30GB",
   "model": "gemma-4-31B-it-Q4_K_M.gguf",
   "model_size": null,
   "quant": null,
   "runtime": "llama.cpp",
   "runtime_version": null,
   "workload": "chat-short",
   "suite_version": "suite-v1",
   "decode_tps": 67.54,
   "prefill_tps": null,
   "ttft_ms": 182.9,
   "prompt_tokens": 0,
   "output_tokens": 253,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_-1k-8qr_v8a",
   "run_url": "https://llm-speed.com/r/r_-1k-8qr_v8a",
   "date": "2026-07-01",
   "hardware": "RTX 5090 (32GB) + AMD Ryzen 7 9850X3D 8-Core Processor (8c) + 30GB",
   "model": "Qwen3.6-27B-Q4_K_M.gguf",
   "model_size": null,
   "quant": null,
   "runtime": "llama.cpp",
   "runtime_version": null,
   "workload": "chat-short",
   "suite_version": "suite-v1",
   "decode_tps": 71.26,
   "prefill_tps": null,
   "ttft_ms": 409.7,
   "prompt_tokens": 0,
   "output_tokens": 256,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_wqhqzea_7xy",
   "run_url": "https://llm-speed.com/r/r_wqhqzea_7xy",
   "date": "2026-06-30",
   "hardware": "RTX 5090 (32GB) + AMD Ryzen 7 9850X3D 8-Core Processor (8c) + 30GB",
   "model": "Qwen3.6-27B-Q4_K_M.gguf",
   "model_size": null,
   "quant": null,
   "runtime": "llama.cpp",
   "runtime_version": null,
   "workload": "chat-short",
   "suite_version": "suite-v1",
   "decode_tps": 71.92,
   "prefill_tps": null,
   "ttft_ms": 530.5,
   "prompt_tokens": 0,
   "output_tokens": 256,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_txvkjgoyyxs",
   "run_url": "https://llm-speed.com/r/r_txvkjgoyyxs",
   "date": "2026-06-27",
   "hardware": "M4 Max (40-core GPU) + 128GB unified",
   "model": "qwen3-coder",
   "model_size": "30.5B",
   "quant": "Q4_K_M",
   "runtime": "ollama",
   "runtime_version": "0.30.11",
   "workload": "chat-short",
   "suite_version": "suite-v1",
   "decode_tps": 93.17,
   "prefill_tps": 13.3,
   "ttft_ms": 8270.0,
   "prompt_tokens": 110,
   "output_tokens": 256,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_roktphpc--8",
   "run_url": "https://llm-speed.com/r/r_roktphpc--8",
   "date": "2026-06-27",
   "hardware": "M4 Max (40-core GPU) + 128GB unified",
   "model": "qwen3-coder-bench-32k",
   "model_size": "30.5B",
   "quant": "Q4_K_M",
   "runtime": "ollama",
   "runtime_version": "0.30.11",
   "workload": "concurrent-decode",
   "suite_version": "suite-v1",
   "decode_tps": 109.66,
   "prefill_tps": null,
   "ttft_ms": null,
   "prompt_tokens": 1024,
   "output_tokens": 256,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_roktphpc--8",
   "run_url": "https://llm-speed.com/r/r_roktphpc--8",
   "date": "2026-06-27",
   "hardware": "M4 Max (40-core GPU) + 128GB unified",
   "model": "qwen3-coder-bench-32k",
   "model_size": "30.5B",
   "quant": "Q4_K_M",
   "runtime": "ollama",
   "runtime_version": "0.30.11",
   "workload": "chat-short",
   "suite_version": "suite-v1",
   "decode_tps": 113.35,
   "prefill_tps": 84.12,
   "ttft_ms": 1307.7,
   "prompt_tokens": 110,
   "output_tokens": 256,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_roktphpc--8",
   "run_url": "https://llm-speed.com/r/r_roktphpc--8",
   "date": "2026-06-27",
   "hardware": "M4 Max (40-core GPU) + 128GB unified",
   "model": "qwen3-coder-bench-32k",
   "model_size": "30.5B",
   "quant": "Q4_K_M",
   "runtime": "ollama",
   "runtime_version": "0.30.11",
   "workload": "chat-long",
   "suite_version": "suite-v1",
   "decode_tps": 97.4,
   "prefill_tps": 1342.8,
   "ttft_ms": 2343.6,
   "prompt_tokens": 3147,
   "output_tokens": 571,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_roktphpc--8",
   "run_url": "https://llm-speed.com/r/r_roktphpc--8",
   "date": "2026-06-27",
   "hardware": "M4 Max (40-core GPU) + 128GB unified",
   "model": "qwen3-coder-bench-32k",
   "model_size": "30.5B",
   "quant": "Q4_K_M",
   "runtime": "ollama",
   "runtime_version": "0.30.11",
   "workload": "agent-trace",
   "suite_version": "suite-v1",
   "decode_tps": 103.15,
   "prefill_tps": 3371.6,
   "ttft_ms": 477.5,
   "prompt_tokens": 3247,
   "output_tokens": 840,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_r0di2hkku1h",
   "run_url": "https://llm-speed.com/r/r_r0di2hkku1h",
   "date": "2026-06-27",
   "hardware": "M4 Max (40-core GPU) + 128GB unified",
   "model": "qwen3-coder",
   "model_size": "30.5B",
   "quant": "Q4_K_M",
   "runtime": "ollama",
   "runtime_version": "0.30.11",
   "workload": "concurrent-decode",
   "suite_version": "suite-v1",
   "decode_tps": 109.88,
   "prefill_tps": null,
   "ttft_ms": null,
   "prompt_tokens": 1024,
   "output_tokens": 256,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_r0di2hkku1h",
   "run_url": "https://llm-speed.com/r/r_r0di2hkku1h",
   "date": "2026-06-27",
   "hardware": "M4 Max (40-core GPU) + 128GB unified",
   "model": "qwen3-coder",
   "model_size": "30.5B",
   "quant": "Q4_K_M",
   "runtime": "ollama",
   "runtime_version": "0.30.11",
   "workload": "chat-short",
   "suite_version": "suite-v1",
   "decode_tps": 94.51,
   "prefill_tps": 65.06,
   "ttft_ms": 1690.7,
   "prompt_tokens": 110,
   "output_tokens": 256,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_r0di2hkku1h",
   "run_url": "https://llm-speed.com/r/r_r0di2hkku1h",
   "date": "2026-06-27",
   "hardware": "M4 Max (40-core GPU) + 128GB unified",
   "model": "qwen3-coder",
   "model_size": "30.5B",
   "quant": "Q4_K_M",
   "runtime": "ollama",
   "runtime_version": "0.30.11",
   "workload": "chat-long",
   "suite_version": "suite-v1",
   "decode_tps": 97.35,
   "prefill_tps": 1397.26,
   "ttft_ms": 2252.3,
   "prompt_tokens": 3147,
   "output_tokens": 574,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_40h1dznuznk",
   "run_url": "https://llm-speed.com/r/r_40h1dznuznk",
   "date": "2026-06-27",
   "hardware": "M4 Max (40-core GPU) + 128GB unified",
   "model": "qwen3-coder",
   "model_size": "30.5B",
   "quant": "Q4_K_M",
   "runtime": "ollama",
   "runtime_version": "0.30.11",
   "workload": "concurrent-decode",
   "suite_version": "suite-v1",
   "decode_tps": 109.11,
   "prefill_tps": null,
   "ttft_ms": null,
   "prompt_tokens": 1024,
   "output_tokens": 256,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_40h1dznuznk",
   "run_url": "https://llm-speed.com/r/r_40h1dznuznk",
   "date": "2026-06-27",
   "hardware": "M4 Max (40-core GPU) + 128GB unified",
   "model": "qwen3-coder",
   "model_size": "30.5B",
   "quant": "Q4_K_M",
   "runtime": "ollama",
   "runtime_version": "0.30.11",
   "workload": "chat-short",
   "suite_version": "suite-v1",
   "decode_tps": 92.7,
   "prefill_tps": 718.97,
   "ttft_ms": 153.0,
   "prompt_tokens": 110,
   "output_tokens": 256,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_40h1dznuznk",
   "run_url": "https://llm-speed.com/r/r_40h1dznuznk",
   "date": "2026-06-27",
   "hardware": "M4 Max (40-core GPU) + 128GB unified",
   "model": "qwen3-coder",
   "model_size": "30.5B",
   "quant": "Q4_K_M",
   "runtime": "ollama",
   "runtime_version": "0.30.11",
   "workload": "chat-long",
   "suite_version": "suite-v1",
   "decode_tps": 94.61,
   "prefill_tps": 1396.52,
   "ttft_ms": 2253.5,
   "prompt_tokens": 3147,
   "output_tokens": 634,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_1o8q4lhgj88",
   "run_url": "https://llm-speed.com/r/r_1o8q4lhgj88",
   "date": "2026-06-27",
   "hardware": "M4 Max (40-core GPU) + 128GB unified",
   "model": "qwen3-coder",
   "model_size": "30.5B",
   "quant": "Q4_K_M",
   "runtime": "ollama",
   "runtime_version": "0.30.11",
   "workload": "concurrent-decode",
   "suite_version": "suite-v1",
   "decode_tps": 67.01,
   "prefill_tps": null,
   "ttft_ms": null,
   "prompt_tokens": 1024,
   "output_tokens": 256,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_1o8q4lhgj88",
   "run_url": "https://llm-speed.com/r/r_1o8q4lhgj88",
   "date": "2026-06-27",
   "hardware": "M4 Max (40-core GPU) + 128GB unified",
   "model": "qwen3-coder",
   "model_size": "30.5B",
   "quant": "Q4_K_M",
   "runtime": "ollama",
   "runtime_version": "0.30.11",
   "workload": "chat-short",
   "suite_version": "suite-v1",
   "decode_tps": 55.45,
   "prefill_tps": 675.16,
   "ttft_ms": 162.9,
   "prompt_tokens": 110,
   "output_tokens": 256,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_1o8q4lhgj88",
   "run_url": "https://llm-speed.com/r/r_1o8q4lhgj88",
   "date": "2026-06-27",
   "hardware": "M4 Max (40-core GPU) + 128GB unified",
   "model": "qwen3-coder",
   "model_size": "30.5B",
   "quant": "Q4_K_M",
   "runtime": "ollama",
   "runtime_version": "0.30.11",
   "workload": "chat-long",
   "suite_version": "suite-v1",
   "decode_tps": 57.86,
   "prefill_tps": 28422.38,
   "ttft_ms": 110.7,
   "prompt_tokens": 3147,
   "output_tokens": 679,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_1pl79r50ofy",
   "run_url": "https://llm-speed.com/r/r_1pl79r50ofy",
   "date": "2026-05-12",
   "hardware": "M3 Ultra (60-core GPU) + 96GB unified",
   "model": "lmstudio-community-Qwen3-Next-80B-A3B-Instruct-MLX-4bit",
   "model_size": "80B",
   "quant": "4bit",
   "runtime": "mlx",
   "runtime_version": "0.31.3",
   "workload": "concurrent-decode",
   "suite_version": "suite-v1",
   "decode_tps": 78.6,
   "prefill_tps": null,
   "ttft_ms": null,
   "prompt_tokens": 1024,
   "output_tokens": 256,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_1pl79r50ofy",
   "run_url": "https://llm-speed.com/r/r_1pl79r50ofy",
   "date": "2026-05-12",
   "hardware": "M3 Ultra (60-core GPU) + 96GB unified",
   "model": "lmstudio-community-Qwen3-Next-80B-A3B-Instruct-MLX-4bit",
   "model_size": "80B",
   "quant": "4bit",
   "runtime": "mlx",
   "runtime_version": "0.31.3",
   "workload": "chat-short",
   "suite_version": "suite-v1",
   "decode_tps": 80.34,
   "prefill_tps": 24.49,
   "ttft_ms": 4492.5,
   "prompt_tokens": 110,
   "output_tokens": 256,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_1pl79r50ofy",
   "run_url": "https://llm-speed.com/r/r_1pl79r50ofy",
   "date": "2026-05-12",
   "hardware": "M3 Ultra (60-core GPU) + 96GB unified",
   "model": "lmstudio-community-Qwen3-Next-80B-A3B-Instruct-MLX-4bit",
   "model_size": "80B",
   "quant": "4bit",
   "runtime": "mlx",
   "runtime_version": "0.31.3",
   "workload": "chat-long",
   "suite_version": "suite-v1",
   "decode_tps": 77.63,
   "prefill_tps": 1608.6,
   "ttft_ms": 1956.4,
   "prompt_tokens": 3147,
   "output_tokens": 819,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_1pl79r50ofy",
   "run_url": "https://llm-speed.com/r/r_1pl79r50ofy",
   "date": "2026-05-12",
   "hardware": "M3 Ultra (60-core GPU) + 96GB unified",
   "model": "lmstudio-community-Qwen3-Next-80B-A3B-Instruct-MLX-4bit",
   "model_size": "80B",
   "quant": "4bit",
   "runtime": "mlx",
   "runtime_version": "0.31.3",
   "workload": "agent-trace",
   "suite_version": "suite-v1",
   "decode_tps": 78.41,
   "prefill_tps": 1586.28,
   "ttft_ms": 1318.2,
   "prompt_tokens": 3247,
   "output_tokens": 840,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_bqsunbd6xa8",
   "run_url": "https://llm-speed.com/r/r_bqsunbd6xa8",
   "date": "2026-05-02",
   "hardware": "RTX 5090 (32GB) + AMD Ryzen 7 9850X3D 8-Core Processor (8c) + 30GB",
   "model": "Qwen3.6-27B-Q4_K_M.gguf",
   "model_size": null,
   "quant": null,
   "runtime": "llama.cpp",
   "runtime_version": null,
   "workload": "chat-short",
   "suite_version": "suite-v1",
   "decode_tps": 69.89,
   "prefill_tps": null,
   "ttft_ms": 3995.5,
   "prompt_tokens": 0,
   "output_tokens": 256,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_1pww-w7p8sd",
   "run_url": "https://llm-speed.com/r/r_1pww-w7p8sd",
   "date": "2026-05-02",
   "hardware": "RTX 5090 (32GB) + AMD Ryzen 7 9850X3D 8-Core Processor (8c) + 30GB",
   "model": "Qwen3.6-27B-Q4_K_M.gguf",
   "model_size": null,
   "quant": null,
   "runtime": "llama.cpp",
   "runtime_version": null,
   "workload": "chat-short",
   "suite_version": "suite-v1",
   "decode_tps": 67.28,
   "prefill_tps": null,
   "ttft_ms": 552.5,
   "prompt_tokens": 0,
   "output_tokens": 256,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_kj4fh_mmzj9",
   "run_url": "https://llm-speed.com/r/r_kj4fh_mmzj9",
   "date": "2026-05-01",
   "hardware": "RTX 5090 (32GB) + AMD Ryzen 7 9850X3D 8-Core Processor (8c) + 30GB",
   "model": "Qwen3.6-27B-Q4_K_M.gguf",
   "model_size": null,
   "quant": null,
   "runtime": "llama.cpp",
   "runtime_version": null,
   "workload": "chat-short",
   "suite_version": "suite-v1",
   "decode_tps": 47.75,
   "prefill_tps": null,
   "ttft_ms": 2833.3,
   "prompt_tokens": 0,
   "output_tokens": 256,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r__b89kg2iica",
   "run_url": "https://llm-speed.com/r/r__b89kg2iica",
   "date": "2026-05-01",
   "hardware": "RTX 5090 (32GB) + AMD Ryzen 7 9850X3D 8-Core Processor (8c) + 30GB",
   "model": "Qwen3.6-27B-Q4_K_M.gguf",
   "model_size": null,
   "quant": null,
   "runtime": "llama.cpp",
   "runtime_version": null,
   "workload": "chat-short",
   "suite_version": "suite-v1",
   "decode_tps": 39.61,
   "prefill_tps": null,
   "ttft_ms": 226.9,
   "prompt_tokens": 0,
   "output_tokens": 256,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_4u7250hj28o",
   "run_url": "https://llm-speed.com/r/r_4u7250hj28o",
   "date": "2026-05-01",
   "hardware": "RTX 5090 (32GB) + AMD Ryzen 7 9850X3D 8-Core Processor (8c) + 30GB",
   "model": "Qwen3.6-27B-Q4_K_M.gguf",
   "model_size": null,
   "quant": null,
   "runtime": "llama.cpp",
   "runtime_version": null,
   "workload": "chat-short",
   "suite_version": "suite-v1",
   "decode_tps": 45.92,
   "prefill_tps": null,
   "ttft_ms": 3088.8,
   "prompt_tokens": 0,
   "output_tokens": 256,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_y2_5y8oo97d",
   "run_url": "https://llm-speed.com/r/r_y2_5y8oo97d",
   "date": "2026-04-28",
   "hardware": "M3 Ultra (60-core GPU) + 96GB unified",
   "model": "mlx-community/stable-code-instruct-3b-4bit",
   "model_size": null,
   "quant": null,
   "runtime": "mlx",
   "runtime_version": "0.31.3",
   "workload": "chat-short",
   "suite_version": "suite-v1",
   "decode_tps": 192.5,
   "prefill_tps": 560.72,
   "ttft_ms": 226.5,
   "prompt_tokens": 127,
   "output_tokens": 256,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_wsxml_39dh_",
   "run_url": "https://llm-speed.com/r/r_wsxml_39dh_",
   "date": "2026-04-28",
   "hardware": "M3 Ultra (60-core GPU) + 96GB unified",
   "model": "mlx-community/starcoder2-15b-4bit",
   "model_size": null,
   "quant": null,
   "runtime": "mlx",
   "runtime_version": "0.31.3",
   "workload": "chat-short",
   "suite_version": "suite-v1",
   "decode_tps": 64.16,
   "prefill_tps": 220.48,
   "ttft_ms": 489.8,
   "prompt_tokens": 108,
   "output_tokens": 256,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_v4bq1sviz4o",
   "run_url": "https://llm-speed.com/r/r_v4bq1sviz4o",
   "date": "2026-04-28",
   "hardware": "M3 Ultra (60-core GPU) + 96GB unified",
   "model": "mlx-community-Qwen2.5-14B-Instruct-4bit",
   "model_size": "14B",
   "quant": "4bit",
   "runtime": "mlx",
   "runtime_version": "0.31.3",
   "workload": "chat-short",
   "suite_version": "suite-v1",
   "decode_tps": 70.85,
   "prefill_tps": 301.75,
   "ttft_ms": 434.1,
   "prompt_tokens": 131,
   "output_tokens": 256,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_v2pbc0rq2l4",
   "run_url": "https://llm-speed.com/r/r_v2pbc0rq2l4",
   "date": "2026-04-28",
   "hardware": "M3 Ultra (60-core GPU) + 96GB unified",
   "model": "mlx-community/Llama-3.1-8B-Instruct-4bit",
   "model_size": null,
   "quant": null,
   "runtime": "mlx",
   "runtime_version": "0.31.3",
   "workload": "chat-short",
   "suite_version": "suite-v1",
   "decode_tps": 130.18,
   "prefill_tps": 400.34,
   "ttft_ms": 339.7,
   "prompt_tokens": 136,
   "output_tokens": 256,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_uoehjq0nvc0",
   "run_url": "https://llm-speed.com/r/r_uoehjq0nvc0",
   "date": "2026-04-28",
   "hardware": "M3 Ultra (60-core GPU) + 96GB unified",
   "model": "mlx-community/Qwen2.5-Coder-7B-Instruct-4bit",
   "model_size": null,
   "quant": null,
   "runtime": "mlx",
   "runtime_version": "0.31.3",
   "workload": "chat-short",
   "suite_version": "suite-v1",
   "decode_tps": 138.55,
   "prefill_tps": 539.59,
   "ttft_ms": 242.8,
   "prompt_tokens": 131,
   "output_tokens": 256,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_sx3a4y9n-m4",
   "run_url": "https://llm-speed.com/r/r_sx3a4y9n-m4",
   "date": "2026-04-28",
   "hardware": "M3 Ultra (60-core GPU) + 96GB unified",
   "model": "mlx-community/llama-3.3-70b-Instruct-4bit",
   "model_size": null,
   "quant": null,
   "runtime": "mlx",
   "runtime_version": "0.31.3",
   "workload": "chat-short",
   "suite_version": "suite-v1",
   "decode_tps": 16.78,
   "prefill_tps": 25.09,
   "ttft_ms": 5419.9,
   "prompt_tokens": 136,
   "output_tokens": 256,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_sqzp-0rdez-",
   "run_url": "https://llm-speed.com/r/r_sqzp-0rdez-",
   "date": "2026-04-28",
   "hardware": "M3 Ultra (60-core GPU) + 96GB unified",
   "model": "mlx-community/phi-4-4bit",
   "model_size": null,
   "quant": null,
   "runtime": "mlx",
   "runtime_version": "0.31.3",
   "workload": "chat-short",
   "suite_version": "suite-v1",
   "decode_tps": 74.37,
   "prefill_tps": 254.99,
   "ttft_ms": 443.2,
   "prompt_tokens": 113,
   "output_tokens": 256,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_q7t7b7dcuz5",
   "run_url": "https://llm-speed.com/r/r_q7t7b7dcuz5",
   "date": "2026-04-28",
   "hardware": "M3 Pro (18-core GPU) + 36GB unified",
   "model": "mlx-community/gemma-2-9b-it-4bit",
   "model_size": null,
   "quant": null,
   "runtime": "mlx",
   "runtime_version": "0.31.3",
   "workload": "chat-short",
   "suite_version": "suite-v1",
   "decode_tps": 21.97,
   "prefill_tps": 102.5,
   "ttft_ms": 1151.3,
   "prompt_tokens": 118,
   "output_tokens": 256,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_pnrrpcdqfo4",
   "run_url": "https://llm-speed.com/r/r_pnrrpcdqfo4",
   "date": "2026-04-28",
   "hardware": "M3 Pro (18-core GPU) + 36GB unified",
   "model": "mlx-community/Qwen3-32B-4bit",
   "model_size": null,
   "quant": null,
   "runtime": "mlx",
   "runtime_version": "0.31.3",
   "workload": "chat-short",
   "suite_version": "suite-v1",
   "decode_tps": 7.16,
   "prefill_tps": 25.65,
   "ttft_ms": 4287.9,
   "prompt_tokens": 110,
   "output_tokens": 256,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_njgxtgyym1e",
   "run_url": "https://llm-speed.com/r/r_njgxtgyym1e",
   "date": "2026-04-28",
   "hardware": "M3 Ultra (60-core GPU) + 96GB unified",
   "model": "mlx-community-Qwen2.5-32B-Instruct-4bit",
   "model_size": "32B",
   "quant": "4bit",
   "runtime": "mlx",
   "runtime_version": "0.31.3",
   "workload": "chat-short",
   "suite_version": "suite-v1",
   "decode_tps": 34.6,
   "prefill_tps": 141.94,
   "ttft_ms": 922.9,
   "prompt_tokens": 131,
   "output_tokens": 256,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_llzv_g-ymaf",
   "run_url": "https://llm-speed.com/r/r_llzv_g-ymaf",
   "date": "2026-04-28",
   "hardware": "M3 Pro (18-core GPU) + 36GB unified",
   "model": "mlx-community-Qwen2.5-7B-Instruct-4bit",
   "model_size": "7B",
   "quant": "4bit",
   "runtime": "mlx",
   "runtime_version": "0.31.3",
   "workload": "chat-short",
   "suite_version": "suite-v1",
   "decode_tps": 30.52,
   "prefill_tps": 161.91,
   "ttft_ms": 809.1,
   "prompt_tokens": 131,
   "output_tokens": 256,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_l_v1-zq_qaz",
   "run_url": "https://llm-speed.com/r/r_l_v1-zq_qaz",
   "date": "2026-04-28",
   "hardware": "M3 Ultra (60-core GPU) + 96GB unified",
   "model": "mlx-community/DeepSeek-Coder-V2-Lite-Instruct-4bit",
   "model_size": null,
   "quant": null,
   "runtime": "mlx",
   "runtime_version": "0.31.3",
   "workload": "chat-short",
   "suite_version": "suite-v1",
   "decode_tps": 168.34,
   "prefill_tps": 291.46,
   "ttft_ms": 449.5,
   "prompt_tokens": 131,
   "output_tokens": 256,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_l36cijqxq4t",
   "run_url": "https://llm-speed.com/r/r_l36cijqxq4t",
   "date": "2026-04-28",
   "hardware": "M3 Ultra (60-core GPU) + 96GB unified",
   "model": "mlx-community/Qwen2.5-Coder-14B-Instruct-4bit",
   "model_size": null,
   "quant": null,
   "runtime": "mlx",
   "runtime_version": "0.31.3",
   "workload": "chat-short",
   "suite_version": "suite-v1",
   "decode_tps": 70.51,
   "prefill_tps": 306.12,
   "ttft_ms": 427.9,
   "prompt_tokens": 131,
   "output_tokens": 256,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_iz137eqvuzy",
   "run_url": "https://llm-speed.com/r/r_iz137eqvuzy",
   "date": "2026-04-28",
   "hardware": "M3 Ultra (60-core GPU) + 96GB unified",
   "model": "mlx-community/gemma-2-9b-it-4bit",
   "model_size": null,
   "quant": null,
   "runtime": "mlx",
   "runtime_version": "0.31.3",
   "workload": "chat-short",
   "suite_version": "suite-v1",
   "decode_tps": 89.45,
   "prefill_tps": 170.67,
   "ttft_ms": 691.4,
   "prompt_tokens": 118,
   "output_tokens": 256,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_ia73dzeue0b",
   "run_url": "https://llm-speed.com/r/r_ia73dzeue0b",
   "date": "2026-04-28",
   "hardware": "M3 Pro (18-core GPU) + 36GB unified",
   "model": "mlx-community-Qwen2.5-14B-Instruct-4bit",
   "model_size": "14B",
   "quant": "4bit",
   "runtime": "mlx",
   "runtime_version": "0.31.3",
   "workload": "chat-short",
   "suite_version": "suite-v1",
   "decode_tps": 14.41,
   "prefill_tps": 117.86,
   "ttft_ms": 1111.5,
   "prompt_tokens": 131,
   "output_tokens": 256,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_h0-use1ypnb",
   "run_url": "https://llm-speed.com/r/r_h0-use1ypnb",
   "date": "2026-04-28",
   "hardware": "M3 Pro (18-core GPU) + 36GB unified",
   "model": "mlx-community/Llama-3.1-8B-Instruct-4bit",
   "model_size": null,
   "quant": null,
   "runtime": "mlx",
   "runtime_version": "0.31.3",
   "workload": "chat-short",
   "suite_version": "suite-v1",
   "decode_tps": 29.2,
   "prefill_tps": 203.28,
   "ttft_ms": 669.0,
   "prompt_tokens": 136,
   "output_tokens": 256,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_fpsca03u2o_",
   "run_url": "https://llm-speed.com/r/r_fpsca03u2o_",
   "date": "2026-04-28",
   "hardware": "M3 Ultra (60-core GPU) + 96GB unified",
   "model": "mlx-community/Qwen3-Coder-30B-A3B-Instruct-4bit",
   "model_size": null,
   "quant": null,
   "runtime": "mlx",
   "runtime_version": "0.31.3",
   "workload": "chat-short",
   "suite_version": "suite-v1",
   "decode_tps": 112.15,
   "prefill_tps": 203.95,
   "ttft_ms": 539.3,
   "prompt_tokens": 110,
   "output_tokens": 256,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_f4x3xaan2ay",
   "run_url": "https://llm-speed.com/r/r_f4x3xaan2ay",
   "date": "2026-04-28",
   "hardware": "M3 Pro (18-core GPU) + 36GB unified",
   "model": "mlx-community/Qwen2.5-32B-Instruct-4bit",
   "model_size": null,
   "quant": null,
   "runtime": "mlx",
   "runtime_version": "0.31.3",
   "workload": "chat-short",
   "suite_version": "suite-v1",
   "decode_tps": 7.05,
   "prefill_tps": 33.41,
   "ttft_ms": 3921.1,
   "prompt_tokens": 131,
   "output_tokens": 256,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_e3t93rscswq",
   "run_url": "https://llm-speed.com/r/r_e3t93rscswq",
   "date": "2026-04-28",
   "hardware": "M3 Pro (18-core GPU) + 36GB unified",
   "model": "mlx-community-Qwen2.5-7B-Instruct-4bit",
   "model_size": "7B",
   "quant": "4bit",
   "runtime": "mlx",
   "runtime_version": "0.31.3",
   "workload": "chat-short",
   "suite_version": "suite-v1",
   "decode_tps": 15.67,
   "prefill_tps": 54.34,
   "ttft_ms": 2410.7,
   "prompt_tokens": 131,
   "output_tokens": 256,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_bue3bee0gw7",
   "run_url": "https://llm-speed.com/r/r_bue3bee0gw7",
   "date": "2026-04-28",
   "hardware": "M3 Ultra (60-core GPU) + 96GB unified",
   "model": "mlx-community/granite-8b-code-instruct-4bit",
   "model_size": null,
   "quant": null,
   "runtime": "mlx",
   "runtime_version": "0.31.3",
   "workload": "chat-short",
   "suite_version": "suite-v1",
   "decode_tps": 112.27,
   "prefill_tps": 525.71,
   "ttft_ms": 220.7,
   "prompt_tokens": 116,
   "output_tokens": 256,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_anmmc80-aoq",
   "run_url": "https://llm-speed.com/r/r_anmmc80-aoq",
   "date": "2026-04-28",
   "hardware": "M3 Ultra (60-core GPU) + 96GB unified",
   "model": "mlx-community/Qwen3-32B-4bit",
   "model_size": null,
   "quant": null,
   "runtime": "mlx",
   "runtime_version": "0.31.3",
   "workload": "chat-short",
   "suite_version": "suite-v1",
   "decode_tps": 34.41,
   "prefill_tps": 95.12,
   "ttft_ms": 1156.4,
   "prompt_tokens": 110,
   "output_tokens": 256,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_79dvtag5fd_",
   "run_url": "https://llm-speed.com/r/r_79dvtag5fd_",
   "date": "2026-04-28",
   "hardware": "M3 Ultra (60-core GPU) + 96GB unified",
   "model": "mlx-community/Codestral-22B-v0.1-4bit",
   "model_size": null,
   "quant": null,
   "runtime": "mlx",
   "runtime_version": "0.31.3",
   "workload": "chat-short",
   "suite_version": "suite-v1",
   "decode_tps": 47.49,
   "prefill_tps": 205.7,
   "ttft_ms": 559.1,
   "prompt_tokens": 115,
   "output_tokens": 256,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_721b4bls_oq",
   "run_url": "https://llm-speed.com/r/r_721b4bls_oq",
   "date": "2026-04-28",
   "hardware": "M3 Ultra (60-core GPU) + 96GB unified",
   "model": "mlx-community/Qwen2.5-Coder-32B-Instruct-4bit",
   "model_size": null,
   "quant": null,
   "runtime": "mlx",
   "runtime_version": "0.31.3",
   "workload": "chat-short",
   "suite_version": "suite-v1",
   "decode_tps": 34.48,
   "prefill_tps": 144.05,
   "ttft_ms": 909.4,
   "prompt_tokens": 131,
   "output_tokens": 256,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_5r6rhiynenc",
   "run_url": "https://llm-speed.com/r/r_5r6rhiynenc",
   "date": "2026-04-28",
   "hardware": "M3 Ultra (60-core GPU) + 96GB unified",
   "model": "mlx-community-Qwen2.5-7B-Instruct-4bit",
   "model_size": "7B",
   "quant": "4bit",
   "runtime": "mlx",
   "runtime_version": "0.31.3",
   "workload": "chat-short",
   "suite_version": "suite-v1",
   "decode_tps": 139.57,
   "prefill_tps": 190.01,
   "ttft_ms": 689.4,
   "prompt_tokens": 131,
   "output_tokens": 256,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_5c80gthqlh6",
   "run_url": "https://llm-speed.com/r/r_5c80gthqlh6",
   "date": "2026-04-28",
   "hardware": "M3 Ultra (60-core GPU) + 96GB unified",
   "model": "mlx-community/qwen2.5-72b-Instruct-4bit",
   "model_size": null,
   "quant": null,
   "runtime": "mlx",
   "runtime_version": "0.31.3",
   "workload": "chat-short",
   "suite_version": "suite-v1",
   "decode_tps": 16.31,
   "prefill_tps": 23.25,
   "ttft_ms": 5635.3,
   "prompt_tokens": 131,
   "output_tokens": 256,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_3ijun8ltjnb",
   "run_url": "https://llm-speed.com/r/r_3ijun8ltjnb",
   "date": "2026-04-28",
   "hardware": "M3 Ultra (60-core GPU) + 96GB unified",
   "model": "mlx-community/gpt-oss-20b-MXFP4-Q4",
   "model_size": null,
   "quant": null,
   "runtime": "mlx",
   "runtime_version": "0.31.3",
   "workload": "chat-short",
   "suite_version": "suite-v1",
   "decode_tps": 152.67,
   "prefill_tps": 239.86,
   "ttft_ms": 692.1,
   "prompt_tokens": 166,
   "output_tokens": 256,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_3hvui9a1yuc",
   "run_url": "https://llm-speed.com/r/r_3hvui9a1yuc",
   "date": "2026-04-28",
   "hardware": "M3 Ultra (60-core GPU) + 96GB unified",
   "model": "mlx-community/Yi-Coder-9B-Chat-4bit",
   "model_size": null,
   "quant": null,
   "runtime": "mlx",
   "runtime_version": "0.31.3",
   "workload": "chat-short",
   "suite_version": "suite-v1",
   "decode_tps": 103.55,
   "prefill_tps": 390.85,
   "ttft_ms": 307.0,
   "prompt_tokens": 120,
   "output_tokens": 256,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_-w8hnn61va_",
   "run_url": "https://llm-speed.com/r/r_-w8hnn61va_",
   "date": "2026-04-28",
   "hardware": "M3 Pro (18-core GPU) + 36GB unified",
   "model": "mlx-community/phi-4-4bit",
   "model_size": null,
   "quant": null,
   "runtime": "mlx",
   "runtime_version": "0.31.3",
   "workload": "chat-short",
   "suite_version": "suite-v1",
   "decode_tps": 15.61,
   "prefill_tps": 120.72,
   "ttft_ms": 936.0,
   "prompt_tokens": 113,
   "output_tokens": 256,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_79bwm4mq_4l",
   "run_url": "https://llm-speed.com/r/r_79bwm4mq_4l",
   "date": "2026-04-27",
   "hardware": "RTX 5090 (32GB) + AMD Ryzen 7 9850X3D 8-Core Processor (8c) + 30GB",
   "model": "Qwen3.6-27B-Q4_K_M.gguf",
   "model_size": null,
   "quant": null,
   "runtime": "llama.cpp",
   "runtime_version": null,
   "workload": "chat-short",
   "suite_version": "suite-v1",
   "decode_tps": 66.28,
   "prefill_tps": null,
   "ttft_ms": 352.7,
   "prompt_tokens": 0,
   "output_tokens": 256,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_bftqtkilvoe",
   "run_url": "https://llm-speed.com/r/r_bftqtkilvoe",
   "date": "2026-04-26",
   "hardware": "M3 Pro (18-core GPU) + 36GB unified",
   "model": "mlx-community-Qwen2.5-0.5B-Instruct-4bit",
   "model_size": "0.5B",
   "quant": "4bit",
   "runtime": "mlx",
   "runtime_version": "0.31.3",
   "workload": "chat-short",
   "suite_version": "suite-v1",
   "decode_tps": 282.62,
   "prefill_tps": 668.8,
   "ttft_ms": 195.9,
   "prompt_tokens": 131,
   "output_tokens": 256,
   "batch_size": 1,
   "label": "measured"
  },
  {
   "run_id": "r_akcbpx5vcqa",
   "run_url": "https://llm-speed.com/r/r_akcbpx5vcqa",
   "date": "2026-04-26",
   "hardware": "M3 Pro (18-core GPU) + 36GB unified",
   "model": "mlx-community-Qwen2.5-0.5B-Instruct-4bit",
   "model_size": "0.5B",
   "quant": "4bit",
   "runtime": "mlx",
   "runtime_version": "0.31.3",
   "workload": "chat-short",
   "suite_version": "suite-v1",
   "decode_tps": 286.52,
   "prefill_tps": 656.2,
   "ttft_ms": 199.6,
   "prompt_tokens": 131,
   "output_tokens": 256,
   "batch_size": 1,
   "label": "measured"
  }
 ]
}