Skip to content
mlx.app

M1 Ultra · 64 GB

800 GB/s memory bandwidth · 51.2 GB usable for models · 12.8 GB left to macOS

The short answer

34 models run comfortably and 2 more will run tight at 8K context. The largest that fits is Llama 3.3 70B Instruct at q4, generating around 16.3 tokens per second.

macOS12.8 GBweights39.7 GBKV cache2.7 GBheadroom8.8 GB

Every model on this Mac

ModelSizeBest quantVerdictSpeed est.
Qwen2.5 0.5B Instruct490Mbf16FITS · 50.1 GB FREE661 tok/s
BGE-M3567Mbf16FITS · 50.1 GB FREE571 tok/s
Qwen3 0.6B600Mbf16FITS · 49.1 GB FREE540 tok/s
Parakeet TDT 0.6B v2600Mbf16FITS · 50.0 GB FREE540 tok/s
Llama 3.2 1B Instruct1.2Bbf16FITS · 48.5 GB FREE261 tok/s
Whisper Large v31.6Bbf16FITS · 48.1 GB FREE209 tok/s
Qwen3 1.7B1.7Bbf16FITS · 46.9 GB FREE191 tok/s
Llama 3.2 3B Instruct3.2Bbf16FITS · 44.3 GB FREE101 tok/s
Qwen3 4B4Bq8FITS · 45.7 GB FREE152 tok/s
Qwen3 Embedding 4B4Bq8FITS · 45.7 GB FREE152 tok/s
Mistral 7B Instruct v0.37.3Bq8FITS · 42.4 GB FREE83.5 tok/s
Qwen2.5 7B Instruct7.6Bbf16FITS · 35.5 GB FREE42.6 tok/s
InternVL3 8B8Bq8FITS · 42.7 GB FREE76.2 tok/s
Llama 3.1 8B Instruct8.0Bbf16FITS · 34.1 GB FREE40.3 tok/s
Qwen3 8B8.2Bq8FITS · 41.3 GB FREE74.4 tok/s
Qwen2.5-VL 7B Instruct8.3Bq8FITS · 41.9 GB FREE73.5 tok/s
Gemma 2 9B IT9.2Bq8FITS · 41.4 GB FREE66.0 tok/s
Llama 3.2 11B Vision Instruct10.7Bbf16FITS · 29.8 GB FREE30.3 tok/s
Qwen2.5 14B Instruct14.7Bq8FITS · 34.0 GB FREE41.5 tok/s
Qwen3 14B14.8Bq8FITS · 34.1 GB FREE41.2 tok/s
GPT-OSS 20B21Bq8FITS · 28.5 GB FREE169 tok/s
Codestral 22B v0.122.2Bq8FITS · 25.7 GB FREE27.5 tok/s
Mistral Small 24B Instruct 250124Bq8FITS · 24.4 GB FREE25.4 tok/s
Gemma 3 27B IT27Bq8FITS · 22.5 GB FREE22.6 tok/s
Gemma 2 27B IT27.2Bq8FITS · 22.3 GB FREE22.4 tok/s
Qwen3 30B-A3B Instruct 250730.5Bq8FITS · 18.0 GB FREE185 tok/s
Qwen3 Coder 30B-A3B Instruct30.5Bq8FITS · 18.0 GB FREE185 tok/s
Qwen2.5 32B Instruct32.5Bq8FITS · 14.5 GB FREE18.8 tok/s
Qwen2.5 Coder 32B Instruct32.5Bq8FITS · 14.5 GB FREE18.8 tok/s
DeepSeek R1 Distill Qwen 32B32.5Bq8FITS · 14.5 GB FREE18.8 tok/s
Qwen3 32B32.8Bq8FITS · 14.2 GB FREE18.6 tok/s
Qwen2.5-VL 32B Instruct33Bq8FITS · 14.0 GB FREE18.5 tok/s
Mixtral 8x7B Instruct v0.146.7Bq4FITS · 23.9 GB FREE89.3 tok/s
Llama 3.3 70B Instruct70.6Bq4FITS · 8.8 GB FREE16.3 tok/s
Qwen2.5 72B Instruct72.7Bq4TIGHT · 7.6 GB FREE15.8 tok/s
Llama 3.2 90B Vision Instruct88Bq4TIGHT · 1.7 GB FREE13.1 tok/s
GPT-OSS 120B116.8BOVER
Qwen3 235B-A22B235BOVER
DeepSeek V3671BOVER