qwen3.6-agent-q6k-ctx128k

on M5 Max

88.3

tok/sdecode

M5 Max (40-core GPU) + 48GB unified

rank in tier

1/1M5 Max runs

best workload

chat-short

where the rig flew

slowest workload

—

single-workload run

backend

ollamaollama is llama.cpp under the hood; pure llama.cpp tends to nudge a few % faster.

faster than

Nemotron-3.5-Bolt-74B-A9B-24GB on RTX 309087.8 tok/s
Qwen3.8-27B-mtp-on on RTX 409087.0 tok/s
Qwen3-Next-80B-A3B-Instruct-MLX-4bit on M3 Ultra80.3 tok/s

This is a shareable view of one signed llm-speed suite-suite-v1 run. Numbers are not edited. The raw record has every workload, the public key, and the fingerprint hash.