MLX on the M2 Max 64GB
llamaperf has 14 MLX reports on the M2 Max 64GB, none of them a plain single-card run with a model size and quant, so there is no median to quote. The reports are listed below with how each one ran.
14 MLX reports on the M2 Max 64GB
Newest first. A run on several cards, with speculative decoding, with part of the model in system RAM or with batched requests says so, since each describes a different setup.
- Nemotron 3 · 4-bit (source config)93.0 t/s
- Qwen 2.5 Coder · 4-bit (source config)18.0 t/s
- Qwen3.5 · 4-bit (source config)88.9 t/s
- Qwen3.5 · 4-bit (source config)20.7 t/s
- GLM-4.7 · 4-bit (source config)52.6 t/s
- Gemma 4 · 4-bit mixed (config)61.6 t/s
- LFM2 · 4-bit mixed (config)117.8 t/s
- Model not identified · 4-bit (source config)24.4 t/s
- gpt-oss · MXFP4-Q4 mixed (config)85.5 t/s
- Qwen3.5 · 4-bit (source config)62.7 t/s
- Llama 3.1 · 4-bit (source config)72.9 t/s
- Granite 3.2 · 4-bit (source config)64.4 t/s
- Qwen2.5 · 4-bit (source config)76.8 t/s
- Ministral 3 · 4-bit (source config)128.3 t/s
Frequently asked
How fast is MLX on the M2 Max 64GB?
llamaperf has 14 MLX reports on the M2 Max 64GB, none of them a plain single-card run with a model size and quant, so there is no median to quote. The reports are listed below with how each one ran.