Benchmark result
Qwen3.8-Flash-Next-oQ4e-mtp-BF16PLE-v1 on Apple M4 Max — 51.4 tok/s
Measured with oMLX 0.31.3 on September 13, 2026.
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; the frame only adds a line that tells this page how tall it is.
How this run compares
17th fastest of 21 runs of this model on Apple M4 Max · median 52.7 tok/s.
sort
- 157.9tok/s78.4oMLX · KV none
- 257.9tok/s74.5oMLX · KV none
- 357.1tok/s85.1oMLX · KV none
- 456.9tok/s83.9oMLX · KV none
- 556.7tok/s85.6oMLX · KV none
- 656.3tok/s83.7oMLX · KV none
- 756.2tok/s64.1oMLX · KV none
- 856.0tok/s84.1oMLX · KV none
- 955.3tok/s83.1oMLX · KV none
- 1053.2tok/s85.2oMLX · KV none
- 1152.7tok/s84.5oMLX · KV none
- 1252.5tok/s82.2oMLX · KV none
- 1351.6tok/s77.5oMLX · KV none
- 1451.5tok/s83.5oMLX · KV none
- 1551.5tok/s81.2oMLX · KV none
- 1651.5tok/s84.1oMLX · KV none
- 1751.4tok/s81.0oMLX · KV nonethis run
- 1851.4tok/s86.4oMLX · KV none
- 1950.8tok/s84.0oMLX · KV none
- 2050.5tok/s81.2oMLX · KV none
- 2150.2tok/s86.4oMLX · KV none
Reproduce this run
tooloMLX 0.31.3
modelQwen3.8-Flash-Next-oQ4e-mtp-BF16PLE-v1
context131,072
kv cachenone
thinkingon
effortxhigh
think limitnone
samplingtemperature 1 · top_p 0.95 · top_k 20 · min_p 0 · presence_penalty 0 · repetition_penalty 1
clientv0.4.59+97
Set those in oMLX, then:
llm-benchmark benchmark --model "Qwen3.8-Flash-Next-oQ4e-mtp-BF16PLE-v1" --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.