What can the Mac mini M4 Pro 24 GB run?
| Text models: 246 | Visual models: 12 | Strongest first |
- Memory
- 24 GB unified
- Bandwidth
- 273 GB/s
- Maker
- Apple
Text models that run on it
| Model | Best fit | Context | Speed | Strongest at |
|---|---|---|---|---|
| Qwen3.8 27B | 3-bit | 24K | 16 tok/s | Chat, Coding, Agents, Docs |
| Muse Glimmer 30B | 3-bit | 56K | 15 tok/s | Docs |
| Gemma 4 26B A4B | 3-bit | 240K | 80 tok/s | Chat, Coding, Agents, Docs |
| Qwen3.5 9B | 8-bit | 152K | 23 tok/s | Chat, Coding, Agents, Docs |
| Gemma 4 12B | 8-bit | 256K | 17 tok/s | Chat, Coding, Docs |
| Nemotron 3.5 Lightning 30B A3B | 3-bit | 32K | 71 tok/s | Agents, Docs |
| Qwen3.5 4B | 8-bit | 256K | 40 tok/s | Agents, Docs |
| Devstral Small 2 24B | 3-bit | 20K | 18 tok/s | Coding |
| LFM2.5 24B A2B | 4-bit | 56K | 91 tok/s | Agents |
| Mistral Small 3.2 24B | 3-bit | 20K | 18 tok/s | Chat, Coding, Agents |
| Ministral 3 14B | 6-bit | 24K | 19 tok/s | Coding, Agents, Docs |
| Gemma 4 E4B | 8-bit | 128K | 36 tok/s | Chat, Coding, Docs |
| gpt-oss 20B | MXFP4 | 128K | 67 tok/s | Agents |
| Ministral 3 8B | 8-bit | 40K | 24 tok/s | Docs |
| Gemma 4 E2B | 8-bit | 128K | 63 tok/s | Chat |
| Nemotron 3 Nano 4B | 8-bit | 256K | 48 tok/s | Docs |
| Llama 3.1 8B | 8-bit | 52K | 26 tok/s | Chat, Coding, Agents, Docs |
| Qwen3 0.6B | 8-bit | 40K | 172 tok/s | Not rated |
| Qwen3 14B | 6-bit | 20K | 18 tok/s | Not rated |
| Qwen3 VL 8B Instruct | 8-bit | 40K | 24 tok/s | Text and images |
| Qwen3 8B | 8-bit | 40K | 26 tok/s | Not rated |
| Qwen2.5 7B Instruct | 8-bit | 32K | 28 tok/s | Not rated |
| Qwen2.5 VL 7B Instruct | 8-bit | 112K | 26 tok/s | Text and images |
| Unlimited OCR | 8-bit | 32K | 141 tok/s | Reading documents |
| Qwen3 VL 4B Instruct | 8-bit | 72K | 42 tok/s | Text and images |
| Qwen3 1.7B | 8-bit | 40K | 84 tok/s | Not rated |
| Qwen3 4B | 8-bit | 40K | 46 tok/s | Not rated |
| Qwen3 4B Instruct 2507 | 8-bit | 72K | 46 tok/s | Not rated |
| Qwen2.5 3B Instruct | 8-bit | 32K | 57 tok/s | Not rated |
| Qwen3.6 27B | 3-bit | 24K | 16 tok/s | Text and images |
| DeepSeek OCR | 8-bit | 8K | 141 tok/s | Reading documents |
| Qwen2.5 Coder 14B Instruct | 6-bit | 16K | 18 tok/s | Coding |
| Qwen2.5 VL 3B Instruct | 8-bit | 125K | 48 tok/s | Text and images |
| Qwen3.5 27B | 3-bit | 24K | 16 tok/s | Text and images |
| NVIDIA Nemotron 3 Nano 30B A3B BF16 | 3-bit | 32K | 71 tok/s | Not rated |
| Qwen2.5 Coder 7B Instruct | 8-bit | 32K | 28 tok/s | Coding |
| Mistral 7B Instruct V0.2 | 8-bit | 32K | 28 tok/s | Not rated |
| Chandra OCR 2 | 8-bit | 256K | 36 tok/s | Reading documents |
| Qwen3 VL 2B Instruct | 8-bit | 112K | 81 tok/s | Text and images |
| MiniCPM5 2B | 8-bit | 128K | 71 tok/s | Not rated |
| Gemma 3 4B IT | 8-bit | 128K | 50 tok/s | Text and images |
| Llama 3.2 3B Instruct | 8-bit | 104K | 57 tok/s | Not rated |
| Qwen2.5 14B Instruct | 6-bit | 16K | 18 tok/s | Not rated |
| Qwen3 1.7B Base | 8-bit | 32K | 96 tok/s | Not rated |
| Diffusiongemma 26B A4B IT | 3-bit | 120K | 77 tok/s | Text and images |
| HunyuanOCR | 8-bit | 128K | 138 tok/s | Text and images |
| Qwen3 0.6B Base | 8-bit | 32K | 197 tok/s | Not rated |
| Qwen2 VL 7B Instruct | 8-bit | 32K | 26 tok/s | Text and images |
| Ornith 1.5 9B | 8-bit | 144K | 23 tok/s | Not rated |
| Unlimited OCR 2 | 8-bit | 128K | 140 tok/s | Reading documents |
| Granite 4.1 3B | 8-bit | 128K | 52 tok/s | Not rated |
| Mistral 7B V0.1 | 8-bit | 32K | 28 tok/s | Not rated |
| Qwen2.5 Coder 3B Instruct | 8-bit | 32K | 57 tok/s | Coding |
| Meta Llama 3 8B Instruct Abliterated V3 | 8-bit | 8K | 26 tok/s | Not rated |
| EXAONE 3.5 7.8B Instruct | 8-bit | 32K | 27 tok/s | Not rated |
| Saiga Llama3 8B | 8-bit | 8K | 26 tok/s | Not rated |
| DeepSeek R1 0528 Qwen3 8B | 8-bit | 44K | 26 tok/s | Not rated |
| Gemma 3 12B IT | 8-bit | 28K | 18 tok/s | Text and images |
| Phi 3.5 Mini Instruct | 8-bit | 28K | 49 tok/s | Not rated |
| Qwen3 4B Thinking 2507 | 8-bit | 72K | 46 tok/s | Not rated |
Best fit is the highest quality, up to 8-bit, that leaves the use case’s context. Ranked by our use-case scores, then by Hugging Face downloads. Speeds are rough, plus or minus 25%. The top 60 of 246 are shown.
Image and video models
| Model | Best fit | Per output | Strongest at |
|---|---|---|---|
| Stable Diffusion XL 1.0 | 16-bit | 80 s | Images |
| Z-Image Turbo | 8-bit | 1.9 min | Images |
| FLUX.1 [dev] | 5-bit | 4.5 min | Images |
| Qwen-Image 2.1 | 4-bit | 37 min | Images |
| FLUX.3 Klein | 8-bit | 2.9 min | Images |
| FLUX.2 [klein] 9B | 4-bit | 63 s | Images |
| Krea 2 Turbo | 5-bit | 4.4 min | Images |
| Stable Diffusion 3.5 Large | 5-bit | 19 min | Images |
| MOSS-Image | 8-bit | 21 min | Images |
| Wan 2.2 TI2V-5B | 8-bit | 3.0 h | Video |
| MOSS Image Base | 8-bit | 21 min | Images |
| HunyuanVideo 1.5 | 4-bit | 15.5 h | Video |
Only just too big
- Qwen3.6 35B A3B
- Gemma 4 31B
- GLM-4.7 Flash
- Olmo 3.1 32B
- Qwen3 32B
- Qwen3 Coder 30B A3B Instruct
- Qwen3.5 35B A3B
- Qwen3 30B A3B
- Qwen3 30B A3B Instruct 2507
- Laguna XS.2
- OTel 2.0 LLM 31B IT
- Qwen2.5 32B Instruct
- MiniCPM V 4.7 35B A3B
- Laguna XS 2.1
- Qwen2.5 VL 32B Instruct
- Qwen2.5 Coder 32B Instruct
- Qwen3 VL 30B A3B Instruct
- DeepSeek R1 Distill Qwen 32B
- Ornith 1.5 35B A3B
- Qwen3 VL 32B Instruct
- Wan2.2 I2V A14B
- Wan2.1 I2V 14B
- Wan 2.1 T2V-14B
- Qwen3 30B A3B Thinking 2507
- Gnani Evon V3.3 30B A3B
- EXAONE 3.5 32B Instruct
- Flint
- Qwen1.5 32B Chat
- Seed OSS 36B Instruct
- Granite 4.2 30B
These need up to a quarter more memory than the Mac mini M4 Pro 24 GB has. Spilling into system RAM or a lower precision can get them running, slowly.
Questions
What is the best LLM for the Mac mini M4 Pro 24 GB?
Of the open models we track, the strongest that run well on the Mac mini M4 Pro 24 GB are Qwen3.8 27B, Muse Glimmer 30B, Gemma 4 26B A4B. 246 text models fit it in all.
Calculated from published specs, not benchmarked. All machines