Benchmark result
LFM2.5-24B-A2B-MLX-8bit on Apple M5 Max — 126.3 tok/s
Measured with oMLX 0.7.1.dev1 on October 15, 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
3rd fastest of 5 runs of this model on Apple M5 Max · median 126.3 tok/s.
sort
- 1127.6tok/s73.0oMLX · 8bit · KV none
- 2127.1tok/s59.0oMLX · 8bit · KV none
- 3126.3tok/s71.5oMLX · 8bit · KV nonethis run
- 4125.7tok/s64.1oMLX · 8bit · KV none
- 5107.6tok/s67.2oMLX · 8bit · KV none
Reproduce this run
tooloMLX 0.7.1.dev1
modelLFM2.5-24B-A2B-MLX-8bit (8bit)
context262,144
kv cachenone
thinkingon
think limitnone
samplingtemperature 1 · top_p 0.95 · top_k 20 · repetition_penalty 1
clientv0.4.96+102
Set those in oMLX, then:
llm-benchmark benchmark --model "LFM2.5-24B-A2B-MLX-8bit" --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.