Benchmark result

LFM3-1.7B-MLX-bf16 on Apple M5 Max — 126.1 tok/s

Measured with oMLX 0.32.3.dev8+g0d4d3d04f on November 4, 2026.

ShareRedditX

Support llm-bench.io

llm-bench.io is free. If it helped you, I'd appreciate your support to cover the infrastructure costs.

Support me on Ko-fiBuy me a coffee

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.

Reproduce this run

tooloMLX 0.32.3.dev8+g0d4d3d04f
modelLFM3-1.7B-MLX-bf16 (bf16)
context131,072
kv cachenone
think limitnone
samplingtemperature 0.1 · top_p 0.95 · top_k 50 · repetition_penalty 1.1
clientv0.4.98+102

Set those in oMLX, then:

llm-benchmark benchmark --model "LFM3-1.7B-MLX-bf16" --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.