Benchmark result

Qwen3.8-Flash-Next-oQ4e-mtp on Apple M2 Ultra — 27.2 tok/s

Measured with oMLX 0.31.3 on September 1, 2026.

ShareReddit X

What the model built

Open full screen →

The coding scenario asks for a playable game in a single HTML file. This is exactly what the model returned, unedited.

How this run compares

1st fastest of 5 runs of this model on Apple M2 Ultra · median 25.8 tok/s.

sort

127.2tok/s81.2oMLXthis run
225.9tok/s78.0oMLX
325.8tok/s81.7oMLX
423.8tok/s84.4oMLX
523.4tok/s79.1oMLX

Compare the nearest 5 →

Qwen3.8-Flash-Next-oQ4e-mtp on other hardware

132.5tok/s79.2Apple M5 Ultra · oQ4e · KV none
75.5tok/s79.5Apple M4 Max · oQ4e · KV none
75.5tok/s83.2Apple M4 Max · oQ4e · KV none
75.4tok/s82.2Apple M4 Max · oQ4e · KV none
75.2tok/s77.9Apple M4 Max · oQ4e · KV none

Reproduce this run

tooloMLX 0.31.3

modelQwen3.8-Flash-Next-oQ4e-mtp

context131,072

thinkingon

clientv0.4.46+97

Set those in oMLX, then:

llm-benchmark benchmark --model "Qwen3.8-Flash-Next-oQ4e-mtp" --tool "oMLX"

The client prompts for context length and thinking mode, and for the KV cache dtype on Unsloth Studio. Sampling is left at the model default — the values above are what the tool reported using, not overrides the client sent.