Can an M3 Ultra run Qwen2.5-72B-Instruct?
llm-speed has not measured Qwen2.5-72B-Instruct on M3 Ultra itself yet, so there is no llm-speed verdict. 1 community-submitted run exists (fastest 16.3 tok/s at 4-bit on mlx); community-submitted, not measured by llm-speed, so it gives no verdict.
1 signed run on llm-speed. Single-request decode speed in tokens per second; the short-chat workload unless noted. This run is community-submitted (signed on the submitter's own machine, not measured by llm-speed, and not used for llm-speed verdicts). Verdicts and headline numbers use only llm-speed's own runs and fit tests.
Measured tok/s by quant and runtime
| Quant | Runtime | Build | Decode tok/s (best run) | Workload | Runs | Source |
|---|---|---|---|---|---|---|
| 4-bit | mlx | 0.31.3 | 16.3 | chat-short | 1 | Community-submitted |
llm-speed has not measured this pair itself, so there is no llm-speed verdict or fastest quant. The rows above are community-submitted, not measured by llm-speed.
VRAM: not recorded for these runs. The VRAM calculator gives an estimate (predicted, not measured).
Every measured row
| Run | Date (UTC) | Workload | Quant | Runtime / build | Prompt tokens | Output tokens | Decode tok/s | Prefill tok/s | TTFT ms | VRAM GiB | Source |
|---|---|---|---|---|---|---|---|---|---|---|---|
| r_5c80gthqlh6 | 2026-04-28 | chat-short | 4-bit | mlx · 0.31.3 | 131 | 256 | 16.3 | 23.2 | 5635 | – | Community-submitted |
Questions
Can an M3 Ultra run Qwen2.5-72B-Instruct?
llm-speed has not measured Qwen2.5-72B-Instruct on M3 Ultra itself yet, so there is no llm-speed verdict. 1 community-submitted run exists (fastest 16.3 tok/s at 4-bit on mlx); community-submitted, not measured by llm-speed, so it gives no verdict.
What did community runs measure for Qwen2.5-72B-Instruct on an M3 Ultra?
1 community-submitted run (community-submitted, not measured by llm-speed): 4-bit on mlx: 16.3 tok/s. They are listed for reference and give no llm-speed verdict.