gpt-oss 120B on the Apple M2 Ultra
Best measured: 85.6 tokens/s generation and 1,649 tokens/s prompt processing (MXFP4, Metal, FA on).
For comparison, our size-scaling estimate for this combination is 117 tok/s.
Every run
| Quant | Backend | Generate tok/s | Prompt tok/s | Test | Build / by / notes | Source |
|---|---|---|---|---|---|---|
| MXFP4 | Metal FA on | 85.6 | 1,649 | tg32 | ggml-org maintainers Benchmarked repo: https://huggingface.co/ggml-org/gpt-oss-120b-GGUF; chip named only by the file's directory (M2 Ultra), GPU core count not stated | llama.cpp benches/ (maintainer runs) |
tg128 = generating 128 tokens; pp512 = reading a 512-token prompt (llama-bench). LocalScore averages several prompt and output lengths, so its numbers run lower than tg128.
No card yet?
Try it on a rented GPU (RunPod)Affiliate link: if you sign up, RigBench may earn a referral credit. It never affects rankings or numbers.
All results for the Apple M2 Ultra · gpt-oss 120B on other hardware