phi-4-4bit

on M3 Ultra

74.4

tok/sdecode

M3 Ultra (60-core GPU) + 96GB unified

rank in tier

17/26M3 Ultra runstop 65%

best workload

chat-short

where the rig flew

slowest workload

—

single-workload run

backend

mlxMLX is the fastest backend on Apple Silicon for dense models.

faster than

Qwen3.6-27B-Q4_K_M.gguf on RTX 509074.0 tok/s
Qwen2.5-Coder-32B-Instruct on RTX 509071.9 tok/s
Qwen2.5-32B-Instruct on RTX 509071.9 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.