Skip to content
mlx.app

M2 Ultra · 192 GB

800 GB/s memory bandwidth · 163.2 GB usable for models · 28.8 GB left to macOS

The short answer

38 models run comfortably and 0 more will run tight at 8K context. The largest that fits is Qwen3 235B-A22B at q4, generating around 53.0 tokens per second.

macOS28.8 GBweights132.2 GBKV cache1.6 GBheadroom29.4 GB

Every model on this Mac

ModelSizeBest quantVerdictSpeed est.
Qwen2.5 0.5B Instruct490Mbf16FITS · 162.1 GB FREE669 tok/s
BGE-M3567Mbf16FITS · 162.1 GB FREE578 tok/s
Qwen3 0.6B600Mbf16FITS · 161.1 GB FREE547 tok/s
Parakeet TDT 0.6B v2600Mbf16FITS · 162.0 GB FREE547 tok/s
Llama 3.2 1B Instruct1.2Bbf16FITS · 160.5 GB FREE265 tok/s
Whisper Large v31.6Bbf16FITS · 160.1 GB FREE212 tok/s
Qwen3 1.7B1.7Bbf16FITS · 158.9 GB FREE193 tok/s
Llama 3.2 3B Instruct3.2Bbf16FITS · 156.3 GB FREE102 tok/s
Qwen3 4B4Bq8FITS · 157.7 GB FREE154 tok/s
Qwen3 Embedding 4B4Bq8FITS · 157.7 GB FREE154 tok/s
Mistral 7B Instruct v0.37.3Bq8FITS · 154.4 GB FREE84.6 tok/s
Qwen2.5 7B Instruct7.6Bbf16FITS · 147.5 GB FREE43.2 tok/s
InternVL3 8B8Bq8FITS · 154.7 GB FREE77.2 tok/s
Llama 3.1 8B Instruct8.0Bbf16FITS · 146.1 GB FREE40.8 tok/s
Qwen3 8B8.2Bq8FITS · 153.3 GB FREE75.3 tok/s
Qwen2.5-VL 7B Instruct8.3Bq8FITS · 153.9 GB FREE74.4 tok/s
Gemma 2 9B IT9.2Bq8FITS · 153.4 GB FREE66.8 tok/s
Llama 3.2 11B Vision Instruct10.7Bbf16FITS · 141.8 GB FREE30.7 tok/s
Qwen2.5 14B Instruct14.7Bq8FITS · 146.0 GB FREE42.0 tok/s
Qwen3 14B14.8Bq8FITS · 146.1 GB FREE41.7 tok/s
GPT-OSS 20B21Bq8FITS · 140.5 GB FREE172 tok/s
Codestral 22B v0.122.2Bq8FITS · 137.7 GB FREE27.8 tok/s
Mistral Small 24B Instruct 250124Bq8FITS · 136.4 GB FREE25.7 tok/s
Gemma 3 27B IT27Bbf16FITS · 109.2 GB FREE12.1 tok/s
Gemma 2 27B IT27.2Bq8FITS · 134.3 GB FREE22.7 tok/s
Qwen3 30B-A3B Instruct 250730.5Bq8FITS · 130.0 GB FREE187 tok/s
Qwen3 Coder 30B-A3B Instruct30.5Bq8FITS · 130.0 GB FREE187 tok/s
Qwen2.5 32B Instruct32.5Bq8FITS · 126.5 GB FREE19.0 tok/s
Qwen2.5 Coder 32B Instruct32.5Bq8FITS · 126.5 GB FREE19.0 tok/s
DeepSeek R1 Distill Qwen 32B32.5Bq8FITS · 126.5 GB FREE19.0 tok/s
Qwen3 32B32.8Bq8FITS · 126.2 GB FREE18.8 tok/s
Qwen2.5-VL 32B Instruct33Bq8FITS · 126.0 GB FREE18.7 tok/s
Mixtral 8x7B Instruct v0.146.7Bq8FITS · 112.5 GB FREE47.9 tok/s
Llama 3.3 70B Instruct70.6Bq8FITS · 85.5 GB FREE8.7 tok/s
Qwen2.5 72B Instruct72.7Bq8FITS · 83.3 GB FREE8.5 tok/s
Llama 3.2 90B Vision Instruct88Bq4FITS · 113.7 GB FREE13.3 tok/s
GPT-OSS 120B116.8Bq8FITS · 38.5 GB FREE121 tok/s
Qwen3 235B-A22B235Bq4FITS · 29.4 GB FREE53.0 tok/s
DeepSeek V3671BOVER