M3 Max LLM benchmark
No signed M3 Max submissions yet. Check the configuration guidance below, or contribute a reproducible measurement from your own machine.
Specs vendor specs, not measured
- Unified memory (max configuration)
- 128 GB
Plan memory before choosing a model
The catalog lists a family maximum of 128 GB of unified memory. Enter your installed amount; smaller configurations have less space. Model weights, context cache and runtime overhead must share the available memory. Capacity alone does not establish which model will fit or how fast it will run. Use the memory-fit calculator for your model and context as an estimate, then verify with the intended runtime. No signed M3 Max measurement is available yet.
Contribute a signed M3 Max benchmark
Run on YOUR hardware to populate this page: pipx install llm-speed && llm-speed bench
$ pipx install llm-speed && llm-speed bench
Community folklore on M3 Max
25 unverified claims extracted from Reddit/HN comments. Lower trust than signed runs above; every row links to the source.
- communityconfidence 75%
79.60tok/s — Qwen3-8B on M3 Max via mlx bf16
our signed data: M3 Max · Qwen3-8B
“Happy to coordinate if you open-source yours first — no point duplicating. EDIT: Benchmarks complete on M3 Max (128GB). I am getting 79.6 tok/s on Qwen3-8B-bf16 (3.41x speedup) with confirmed bit-for-bit parity against the baseline. For those interested in the MLX-specific …”
- communityconfidence 75%
40.00tok/s — gpt-oss-120B on M3 Max via llama.cpp Q2
our signed data: M3 Max · gpt-oss-120B
“5-air Both are MoE models which improve speed and just about max out the amount of ram you'd have. On my 128GB M3 Max macbook, these are \~40 t/s which is fine (note that glm is slower and both are faster running with mlx instead of llama.cpp). I think the Ryzen is broadly simi…”
- communityconfidence 65%
10.00tok/s — Llama 2 70B on M3 Max Q4_0
our signed data: M3 Max · Llama 2 70B
“OPS, 400GB/s MBW)<p>The results for Llama 2 70B Q4_0 (39GB) was 8.5 tok/s for text generation (you'd expect a theoretical max of a bit over 10 tok/s based on theoretical MBW) and a prompt processing of 19 tok/s. On a 4K context conversation, that means you would be waiting about …”
- communityconfidence 65%
8.50tok/s — Llama 2 70B on M3 Max Q4_0
our signed data: M3 Max · Llama 2 70B
“with their MBP 14 w/ an M3 Max [1] (128GB, 40CU, theoretical: 28.4 FP16 TFLOPS, 400GB/s MBW)<p>The results for Llama 2 70B Q4_0 (39GB) was 8.5 tok/s for text generation (you'd expect a theoretical max of a bit over 10 tok/s based on theoretical MBW) and a prompt processing of 19 …”
- communityconfidence 60%
7.00tok/s — deepseek-r1 on M3 Max via ollama
our signed data: M3 Max · deepseek-r1
“oubles the electricity cost compared to RTX 6000 ADA (19tps) or RTX A6000 (12tps). \- \`M3 Max 40GPU\` has high memory but only delivers 3-7 tps for \`deepseek-r1:70b\`. It is also loud, and the GPU temperature is high (> 90 C). https://preview.redd.it/8r7cwajfn9ee1.png?width=1…”
- communityconfidence 60%
13.27tok/s — gpt-oss-120b on M3 Max via lm-studio
our signed data: M3 Max · gpt-oss-120b
“2 tok/s mradermacher/apriel-1.5-15b-thinker | 0% | M3MAX; 22.91 tok/s; 50 Schema Violation Errors nvidia-nemotron-nano-9b-v2s | 0% | M3MAX; 13.27 tok/s; 50 Incomplete Response Errors openai/gpt-oss-120b | 0% | M3MAX; 26.58 tok/s; 30 Incomplete Response Errors; 9 Schema Violation …”
- communityconfidence 60%
22.91tok/s — gpt-oss-120b on M3 Max via lm-studio
our signed data: M3 Max · gpt-oss-120b
“509 | 100% | 5090; 29.73 tok/s mistralai/magistral-small-2509 | 100% | M3MAX; 15.92 tok/s mradermacher/apriel-1.5-15b-thinker | 0% | M3MAX; 22.91 tok/s; 50 Schema Violation Errors nvidia-nemotron-nano-9b-v2s | 0% | M3MAX; 13.27 tok/s; 50 Incomplete Response Errors openai/gpt-oss-…”
- communityconfidence 60%
15.92tok/s — gpt-oss-120b on M3 Max via lm-studio
our signed data: M3 Max · gpt-oss-120b
“ok/s; 2 Incomplete Response Errors mistralai/magistral-small-2509 | 100% | 5090; 29.73 tok/s mistralai/magistral-small-2509 | 100% | M3MAX; 15.92 tok/s mradermacher/apriel-1.5-15b-thinker | 0% | M3MAX; 22.91 tok/s; 50 Schema Violation Errors nvidia-nemotron-nano-9b-v2s | 0% | M3M…”
- communityconfidence 60%
75.19tok/s — gemma-3-27b on M3 Max via lm-studio
our signed data: M3 Max · gemma-3-27b
“Violation Errors google/gemma-3-27b | 100% | 5090; 51.20 tok/s kat-dev | 100% | 5090; 43.61 tok/s kimi-vl-a3b-thinking-2506 | 96% | M3MAX; 75.19 tok/s; 2 Incomplete Response Errors mistralai/magistral-small-2509 | 100% | 5090; 29.73 tok/s mistralai/magistral-small-2509 | 100% | M…”
- communityconfidence 60%
24.19tok/s — gemma-3-27b on M3 Max via lm-studio
our signed data: M3 Max · gemma-3-27b
“uesting structured output from LM Studio. Results: Model | Pass Percentage | Notes (50 runs per model) :-|:-|:- glm-4.5-air | 86% | M3MAX; 24.19 tok/s; 2 Incomplete Response Errors; 5 Schema Violation Errors google/gemma-3-27b | 100% | 5090; 51.20 tok/s kat-dev | 100% | 5090; 43…”
Request a run on M3 Max
No measured run yet for these pairs. Vote for the one you want; the most-requested runs go first. See all requests.
We never sell or share your email. Privacy