Which AI models can the Apple M4 Pro (24GB) run?
Every model we track, checked against the Apple M4 Pro (24GB) at 8K context. The largest that runs well is gpt-oss 20B.
Unified memory24 GB16.1 GB usable by the GPU
Bandwidth273GB/s
Runs well28/69models at Q4_K_M or better
Fastest model~146tok/s · Qwen3.6 35B-A3B
Every model on the Apple M4 Pro (24GB)
Bars show memory used out of 16.1 GB. Pick a row for the full breakdown.
28 Runs well
NameMemory usedBest quantSpeed
gpt-oss 20B21BQ5_K_M · 15.7 GB~64 tok/s
Qwen3 14B15BQ6_K · 14.1 GB~13 tok/s- Qwen2.5 Coder 14B15BQ6_K · 14.3 GB~13 tok/s
DeepSeek R1 Distill Qwen 14B15BQ6_K · 14.3 GB~13 tok/s- Qwen2.5 14B15BQ6_K · 14.3 GB~14 tok/s
Phi-4 14B15BQ6_K · 14.3 GB~14 tok/s- Phi-4 Reasoning 14B15BQ6_K · 14.3 GB~14 tok/s
Mistral Nemo 12B12BQ8_0 · 14.9 GB~13 tok/s
Gemma 3 12B12BQ8_0 · 14.4 GB~13 tok/s- Gemma 4 12B12BQ8_0 · 13.7 GB~13 tok/s
- Gemma 2 9B9.2BQ8_0 · 12.5 GB~17 tok/s
- Qwen3 8B8.2BQ8_0 · 10.5 GB~19 tok/s
- DeepSeek R1 0528 Qwen3 8B8.2BQ8_0 · 10.5 GB~19 tok/s
Granite 3.3 8B8.2BQ8_0 · 10.6 GB~19 tok/s
Llama 3.1 8B8BQ8_0 · 10.2 GB~19 tok/s- Qwen2.5 7B7.6BQ8_0 · 9.2 GB~20 tok/s
- Qwen2.5 Coder 7B7.6BQ8_0 · 9.2 GB~20 tok/s
- DeepSeek R1 Distill Qwen 7B7.6BQ8_0 · 9.2 GB~20 tok/s
- Mistral 7B v0.37.3BQ8_0 · 9.4 GB~21 tok/s
- Gemma 3 4B4.3BFP16 · 9.5 GB~19 tok/s
- Qwen3 4B4BFP16 · 9.8 GB~20 tok/s
- Phi-4 Mini 3.8B3.8BFP16 · 9.4 GB~21 tok/s
- Phi-3.5 Mini 3.8B3.8BFP16 · 11.5 GB~21 tok/s
- Llama 3.2 3B3.2BFP16 · 8 GB~26 tok/s
- Qwen3 1.7B2BFP16 · 5.6 GB~40 tok/s
- Llama 3.2 1B1.2BFP16 · 3.3 GB~66 tok/s
- Gemma 3 1B1BFP16 · 2.6 GB~82 tok/s
- Qwen3 0.6B0.8BFP16 · 3 GB~109 tok/s
20 Fits with heavy quantization
NameMemory usedBest quantSpeed
- Qwen3.6 35B-A3B36BQ2_K · 14.3 GB~146 tok/s
- Qwen2.5 32B33BQ2_K · 15 GB~13 tok/s
- DeepSeek R1 Distill Qwen 32B33BQ2_K · 15 GB~13 tok/s
- Qwen3 32B33BQ2_K · 15 GB~13 tok/s
- QwQ 32B33BQ2_K · 15 GB~13 tok/s
- Qwen2.5 Coder 32B33BQ2_K · 15 GB~13 tok/s
Nemotron 3 Nano 30B-A3B32BQ2_K · 12.5 GB~125 tok/s- Gemma 4 31B31BQ2_K · 13.5 GB~14 tok/s
GLM-4.7 Flash 30B-A3B31BQ2_K · 12.7 GB~146 tok/s- Qwen3 30B-A3B31BQ2_K · 12.8 GB~132 tok/s
- Qwen3 Coder 30B-A3B31BQ2_K · 12.8 GB~132 tok/s
- Qwen3.8 27B28BQ3_K_M · 14.7 GB~12 tok/s
- Qwen3.6 27B28BQ3_K_M · 14.7 GB~12 tok/s
- Gemma 3 27B27BQ3_K_M · 15.1 GB~12 tok/s
- Gemma 2 27B27BQ2_K · 13.1 GB~16 tok/s
- Gemma 4 26B-A4B26BQ3_K_M · 13.5 GB~88 tok/s
- Mistral Small 3.2 24B24BQ3_K_M · 13.7 GB~14 tok/s
- Mistral Small 24B24BQ3_K_M · 13.5 GB~14 tok/s
- Devstral Small 24B24BQ3_K_M · 13.5 GB~14 tok/s
- Magistral Small 24B24BQ3_K_M · 13.5 GB~14 tok/s
21 Too big
NameMemory usedBest quantSpeed
Kimi K3.1 Pro 2.8T-A104B2800Bneeds 1698.3 GB16.1 GB free- Qwen3.8 2.4T-A95B2446Bneeds 1484.3 GB16.1 GB free
- DeepSeek V4 Pro 1.6T-A49B1600Bneeds 971.2 GB16.1 GB free
- Mistral Large 4 1T-A52B1000Bneeds 607.5 GB16.1 GB free
- DeepSeek V4.1 Flash763Bneeds 463.4 GB16.1 GB free
- GLM-5.3753Bneeds 457.8 GB16.1 GB free
- GLM-5 744B-A40B744Bneeds 452.4 GB16.1 GB free
MiniMax M3 428B-A23B427Bneeds 260.5 GB16.1 GB free- GLM-5.3 Flash 320B-A18B321Bneeds 195.3 GB16.1 GB free
- DeepSeek V4 Flash 284B-A13B284Bneeds 173 GB16.1 GB free
- Qwen3 235B-A22B235Bneeds 144.7 GB16.1 GB free
- Qwen3.8 Flash Next180Bneeds 109.9 GB16.1 GB free
- gpt-oss 120B117Bneeds 71.7 GB16.1 GB free
- Llama 4 Scout 17B-16E109Bneeds 68.3 GB16.1 GB free
- GLM-4.5-Air 106B-A12B106Bneeds 66.4 GB16.1 GB free
- Qwen3 Coder Next 80B-A3B80Bneeds 49.1 GB16.1 GB free
- Qwen2.5 72B73Bneeds 47.4 GB16.1 GB free
- Llama 3.1 70B71Bneeds 46.1 GB16.1 GB free
- Llama 3.3 70B71Bneeds 46.1 GB16.1 GB free
- DeepSeek R1 Distill Llama 70B71Bneeds 46.1 GB16.1 GB free
- Mixtral 8x7B47Bneeds 30 GB16.1 GB free
Run it in the cloud
Need more than 24 GB? Rent a bigger GPU by the hour.
Rent a cloud GPU by the hour. No upfront cost, stop any time.
AMD Radeon RX 7900 XTX 24 GB