Can It Run?
HomeMacs › M3 Pro 18GB

Which AI models run on an M3 Pro 18GB Mac?

M3 Pro · 14-core GPU · 150 GB/s memory bandwidth · MacBook Pro 14" (M3 Pro), MacBook Pro 16" (M3 Pro)

Unified memory
18 GB
soldered, not upgradeable
Usable for a model
15.0 GiB
after macOS takes its share
Bandwidth
150 GB/s
this sets your token speed
Models that fit
19 of 39
comfortably, at Q4_K_M
Short answerThe largest model this machine runs comfortably is DeepSeek-R1-Distill-Qwen-14B — 11.2 GiB loaded, about 13 tokens/sec.

Every model on an M3 Pro 18GB

Assumes Q4_K_M weights and an 8K context with an FP16 KV cache. Click any row for the full breakdown and what to do if it does not fit.

ModelParamsNeedsTok/sVerdict
DeepSeek V4.1 Flash763.21B458.6 GiB0No
DeepSeek R1684.49B411.3 GiB3No
Llama 3.1 405B405B247.3 GiB0No
DeepSeek V4 Flash Vision Exp304.65B183.9 GiB1No
DeepSeek V4 Flash 0731304.18B183.7 GiB1No
DeepSeek V4 Flash290.94B175.7 GiB1No
gpt-oss-120b116.8B71.3 GiB24No
Qwen2.5 72B72.7B46.8 GiB3No
DeepSeek-R1-Distill-Llama-70B70.6B45.6 GiB3No
Llama 3.3 70B70.6B45.6 GiB3No
Mixtral 8x7B46.7B29.8 GiB9No
DeepSeek-R1-Distill-Qwen-32B32.8B22.4 GiB6No
Qwen2.5 32B32.8B22.4 GiB6No
Qwen2.5-Coder 32B32.8B22.4 GiB6No
Qwen3 32B32.8B22.4 GiB6No
Qwen3 30B-A3B30.5B19.8 GiB37No
Gemma 3 27B27.4B21.1 GiB7No
Gemma 2 27B27.2B20.0 GiB7No
Mistral Small 3 24B23.6B16.2 GiB8No
gpt-oss-20b20.9B13.7 GiB34Yes, but tight
DeepSeek-R1-Distill-Qwen-14B14.8B11.2 GiB13Yes, comfortably
Qwen2.5 14B14.8B11.2 GiB13Yes, comfortably
Qwen2.5-Coder 14B14.8B11.2 GiB13Yes, comfortably
Qwen3 14B14.8B10.9 GiB13Yes, comfortably
Phi-4 14B14.7B11.2 GiB13Yes, comfortably
Gemma 3 12B12.2B11.1 GiB16Yes, comfortably
Gemma 2 9B9.24B9.0 GiB21Yes, comfortably
Qwen3 8B8.2B6.8 GiB24Yes, comfortably
Llama 3.1 8B8.03B6.6 GiB24Yes, comfortably
DeepSeek-R1-Distill-Qwen-7B7.62B5.8 GiB26Yes, comfortably
Qwen2.5 7B7.62B5.8 GiB26Yes, comfortably
Qwen2.5-Coder 7B7.62B5.8 GiB26Yes, comfortably
Mistral 7B v0.37.25B6.1 GiB27Yes, comfortably
Gemma 3 4B4.3B4.4 GiB46Yes, comfortably
Llama 3.2 3B3.21B3.6 GiB61Yes, comfortably
Qwen2.5 3B3.09B2.9 GiB63Yes, comfortably
Gemma 2 2B2.61B3.0 GiB75Yes, comfortably
Qwen2.5 1.5B1.54B1.9 GiB127Yes, comfortably
Llama 3.2 1B1.24B1.8 GiB158Yes, comfortably