Benchmark result
Qwen3.8-Flash-Next-oQ3.5e-mtp on Apple M5 Ultra — 88.9 tok/s
Measured with oMLX 0.8.0rc1 on November 13, 2026.
Support llm-bench.io
llm-bench.io is free. If it helped you, I'd appreciate your support to cover the infrastructure costs.
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
5th fastest of 5 runs of this model on Apple M5 Ultra · median 97.2 tok/s.
sort
- 199.6tok/s82.0oMLX · oQ3 · KV none
- 298.7tok/s90.3oMLX · oQ3 · KV none
- 397.2tok/s91.5oMLX · oQ3 · KV none
- 491.6tok/s83.9oMLX · oQ3 · KV none
- 588.9tok/s85.9oMLX · oQ3 · KV nonethis run
Reproduce this run
tooloMLX 0.8.0rc1
modelQwen3.8-Flash-Next-oQ3.5e-mtp (oQ3)
context65,536
kv cachenone
effortxhigh (model default)
think limit8,192 tokens
samplingtemperature 1 · top_p 0.95 · top_k 20 · repetition_penalty 1
clientv0.4.99+102
Set those in oMLX, then:
llm-benchmark benchmark --model "Qwen3.8-Flash-Next-oQ3.5e-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.