Benchmark result
Nex-N2.5-mini-MLX-4bit on Apple M5 Max — 138.4 tok/s
Measured with oMLX 0.31.3 on September 12, 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
7th fastest of 20 runs of this model on Apple M5 Max · median 137.2 tok/s.
sort
- 1140.7tok/s83.5oMLX · KV none
- 2139.6tok/s80.7oMLX · KV none
- 3139.0tok/s78.7oMLX · KV none
- 4138.8tok/s75.8oMLX · KV none
- 5138.8tok/s81.2oMLX · KV none
- 6138.6tok/s80.3oMLX · KV none
- 7138.4tok/s78.7oMLX · KV nonethis run
- 8138.1tok/s83.3oMLX · KV none
- 9137.9tok/s83.5oMLX · KV none
- 10137.4tok/s65.9oMLX · KV none
- 11136.9tok/s81.3oMLX · KV none
- 12136.8tok/s83.1oMLX · KV none
- 13136.7tok/s80.9oMLX · KV none
- 14136.5tok/s82.4oMLX · KV none
- 15135.9tok/s82.8oMLX · KV none
- 16135.7tok/s76.3oMLX · KV none
- 17134.2tok/s81.5oMLX · KV none
- 18133.6tok/s82.9oMLX · KV none
- 19133.2tok/s79.2oMLX · KV none
- 20132.6tok/s85.2oMLX · KV none
Reproduce this run
tooloMLX 0.31.3
modelNex-N2.5-mini-MLX-4bit
context262,144
kv cachenone
thinkingon
efforthigh
think limitnone
samplingtemperature 1 · top_p 0.95 · top_k 20 · repetition_penalty 1
clientv0.4.59+97
Set those in oMLX, then:
llm-benchmark benchmark --model "Nex-N2.5-mini-MLX-4bit" --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.