Qwen3.8 35B (3B active)
on M4 Pro 24GB · oMLX
Summary
User reports Qwen3.8 35B-A3B at an average of around 45 tok/s generated over the past week on a MacBook Pro M4 Pro 24GB.
Setup is oMLX and Inco Splash with IQ3_M weights, after upgrading to 0.7.3 and updating Python.
The 45 tok/s is described as a weekly average rather than a single benchmark run.
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.
Qwen3.8 35B · 3B active at 3-bit by hardware
- 45.0 t/s · one run
Bars are medians on one scale, 0 to 45.0 tokens/s. Press ▶ to watch text arrive at a row's speed.
Models most run on the M4 Pro 24GB
- 45.0 t/s · one run
Bars are medians on one scale, 0 to 45.0 tokens/s. Press ▶ to watch text arrive at a row's speed.
Everything reported on the M4 Pro 24GB →Every Qwen3.8 report →Estimate Qwen3.8 on this card →