Skip to content
llm-speed
← raw run
wrapped·Aug 18, 2026
/r/r_w-p7v--eync
your run

llama3.2

on M1 Pro
103
tok/sdecode
M1 Pro (16-core GPU) + 16GB unified
rank in tier
1/1M1 Pro runs
best workload
chat-short
where the rig flew
slowest workload
chat-long81.3 tok/s
where the rig struggled
backend
ollamaollama is llama.cpp under the hood; pure llama.cpp tends to nudge a few % faster.
faster than
  • Codestral-22B-v0.1 on RTX 5090100 tok/s
  • gemma-2-9b-it-4bit on M3 Ultra89.5 tok/s
  • qwen3.6-agent-q6k-ctx128k on M5 Max88.3 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.