Benchmark result
Swift-1.5-Qwen3.8-27b-oQ8e-mtp on Apple M5 Max — 18.0 tok/s
Measured with oMLX 0.32.0 on September 28, 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.
How this run compares
5th fastest of 5 runs of this model on Apple M5 Max · median 34.5 tok/s.
sort
- 134.8tok/s86.7oMLX · oQ8e · KV none
- 234.8tok/s87.2oMLX · oQ8e · KV none
- 334.5tok/s84.3oMLX · oQ8e · KV none
- 434.1tok/s84.9oMLX · oQ8e · KV none
- 518.0tok/s83.0oMLX · oQ8e · KV nonethis run
Reproduce this run
tooloMLX 0.32.0
modelSwift-1.5-Qwen3.8-27b-oQ8e-mtp (oQ8e)
context262,144
kv cachenone
thinkingon
effortxhigh
samplingtemperature 1 · top_p 0.95 · top_k 20 · repetition_penalty 1
clientv0.4.85+101
Set those in oMLX, then:
llm-benchmark benchmark --model "Swift-1.5-Qwen3.8-27b-oQ8e-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.