Benchmark result
Qwen3.8-Flash-Next-REAP-384-oQ5e-BF16-MTP-PLE on Apple M4 Max — 61.1 tok/s
Measured with oMLX 0.32.0 on September 22, 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
13th fastest of 23 runs of this model on Apple M4 Max · median 61.1 tok/s.
sort
- 162.1tok/s81.5oMLX · oQ5e · KV tq8
- 262.1tok/s81.7oMLX · oQ5e · KV tq8
- 362.0tok/s86.3oMLX · oQ5e · KV tq8
- 461.9tok/s84.3oMLX · oQ5e · KV tq8
- 561.9tok/s85.6oMLX · oQ5e · KV tq8
- 661.9tok/s82.0oMLX · oQ5e · KV tq8
- 761.5tok/s79.1oMLX · oQ5e · KV tq8
- 861.5tok/s84.1oMLX · oQ5e · KV tq8
- 961.5tok/s81.4oMLX · oQ5e · KV tq8
- 1061.4tok/s83.2oMLX · oQ5e · KV tq8
- 1161.1tok/s84.6oMLX · oQ5e · KV tq8
- 1261.1tok/s82.8oMLX · oQ5e · KV tq8
- 1361.1tok/s82.0oMLX · oQ5e · KV tq8this run
- 1461.1tok/s78.8oMLX · oQ5e · KV tq8
- 1560.9tok/s81.0oMLX · oQ5e · KV tq8
- 1660.9tok/s80.8oMLX · oQ5e · KV tq8
- 1760.8tok/s83.7oMLX · oQ5e · KV tq8
- 1860.6tok/s83.3oMLX · oQ5e · KV tq8
- 1960.5tok/s83.7oMLX · oQ5e · KV tq8
- 2060.5tok/s83.4oMLX · oQ5e · KV tq8
- 2160.1tok/s79.5oMLX · oQ5e · KV tq8
- 2259.7tok/s83.0oMLX · oQ5e · KV tq8
- 2358.9tok/s83.6oMLX · oQ5e · KV tq8
Reproduce this run
tooloMLX 0.32.0
modelQwen3.8-Flash-Next-REAP-384-oQ5e-BF16-MTP-PLE (oQ5e)
context262,144
kv cachetq8
effortmedium
think limitnone
samplingtemperature 1 · top_p 0.95 · top_k 20 · repetition_penalty 1
clientv0.4.80+98
Set those in oMLX, then:
llm-benchmark benchmark --model "Qwen3.8-Flash-Next-REAP-384-oQ5e-BF16-MTP-PLE" --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.