Skip to content
llm-speed
← raw run
wrapped·Apr 26, 2026
/r/r_n4t6xaak-jk
your run

deepseek-v3.2

on M3 Pro
20.0
tok/sdecode
M3 Pro (18-core GPU) + 36GB unified
rank in tier
17/27M3 Pro runstop 63%
best workload
chat-short
where the rig flew
slowest workload
—
single-workload run
backend
hosted-apiHosted API — the number is your network + the provider's rig, not yours.
faster than
  • llama-3.3-70b-Instruct-4bit on M3 Ultra16.8 tok/s
  • qwen2.5-72b-Instruct-4bit on M3 Ultra16.3 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.