Use cases
coding
Summary
User reports Mellum2.1-12B-A2.5B-Thinking at 82.4 token/sec on an M1 Max MacBook Pro with 32 GB.
Setup is oMLX with an oQ4e quant.
The figure is described as achieved in normal thinking and coding work; the post shares a benchmark image and mentions running multiple models on the same machine.
Compare this run
Medians of plain runs: one device, one request, no speculative decoding, the whole model in the device's memory. The report above is one setup, so read it beside them and not as one of them.
Mellum2.1 12B · 2.5B active at 4-bit by hardware
- 82.4 t/s · one run
- 64.0 t/s · one run
Bars are medians on one scale, 0 to 82.4 tokens/s. Press ▶ to watch text arrive at a row's speed.
Models most run on the M1 Max 32GB
- 82.4 t/s · one run
- Qwen3.8 27B4-bit15.8 t/s · one run
Bars are medians on one scale, 0 to 82.4 tokens/s. Press ▶ to watch text arrive at a row's speed.
Everything reported on the M1 Max 32GB →Every Mellum2.1 report →Estimate Mellum2.1 on this card →