What can the Mac mini M4 16 GB run?
| Text models: 242 | Visual models: 3 | Strongest first |
- Memory
- 16 GB unified
- Bandwidth
- 120 GB/s
- Maker
- Apple
Text models that run on it
| Model | Best fit | Context | Speed | Strongest at |
|---|---|---|---|---|
| Qwen3.5 9B | 6-bit | 56K | 13 tok/s | Chat, Coding, Agents, Docs |
| Gemma 4 12B | 5-bit | 128K | 11 tok/s | Chat, Coding, Docs |
| Qwen3.5 4B | 8-bit | 144K | 19 tok/s | Agents, Docs |
| Ministral 3 14B | 3-bit | 16K | 14 tok/s | Coding, Agents, Docs |
| Gemma 4 E4B | 8-bit | 80K | 17 tok/s | Chat, Coding, Docs |
| Ministral 3 8B | 6-bit | 20K | 14 tok/s | Docs |
| Gemma 4 E2B | 8-bit | 128K | 31 tok/s | Chat |
| Nemotron 3 Nano 4B | 8-bit | 256K | 23 tok/s | Docs |
| Llama 3.1 8B | 6-bit | 24K | 15 tok/s | Chat, Coding, Agents, Docs |
| Qwen3 0.6B | 8-bit | 40K | 97 tok/s | Not rated |
| Qwen3 14B | 3-bit | 16K | 14 tok/s | Not rated |
| Qwen3 VL 8B Instruct | 6-bit | 16K | 14 tok/s | Text and images |
| Qwen3 8B | 6-bit | 20K | 15 tok/s | Not rated |
| Qwen2.5 7B Instruct | 8-bit | 32K | 13 tok/s | Not rated |
| Qwen2.5 VL 7B Instruct | 8-bit | 20K | 12 tok/s | Text and images |
| Unlimited OCR | 8-bit | 32K | 71 tok/s | Reading documents |
| Qwen2.5 1.5B Instruct | 8-bit | 32K | 53 tok/s | Not rated |
| Qwen3.5 2B | 8-bit | 256K | 37 tok/s | Text and images |
| Qwen3 VL 4B Instruct | 8-bit | 36K | 20 tok/s | Text and images |
| Qwen3 1.7B | 8-bit | 40K | 41 tok/s | Not rated |
| Qwen3 4B | 8-bit | 36K | 22 tok/s | Not rated |
| Qwen3 4B Instruct 2507 | 8-bit | 36K | 22 tok/s | Not rated |
| Qwen2.5 3B Instruct | 8-bit | 32K | 28 tok/s | Not rated |
| DeepSeek OCR | 8-bit | 8K | 71 tok/s | Reading documents |
| Qwen2.5 VL 3B Instruct | 8-bit | 125K | 23 tok/s | Text and images |
| Qwen2.5 Coder 7B Instruct | 8-bit | 32K | 13 tok/s | Coding |
| Mistral 7B Instruct V0.2 | 8-bit | 20K | 13 tok/s | Not rated |
| Chandra OCR 2 | 8-bit | 128K | 17 tok/s | Reading documents |
| Llama 3.2 1B Instruct | 8-bit | 128K | 68 tok/s | Not rated |
| Qwen3 VL 2B Instruct | 8-bit | 64K | 40 tok/s | Text and images |
| Qwen2 VL 2B Instruct | 8-bit | 32K | 38 tok/s | Text and images |
| MiniCPM5 2B | 8-bit | 128K | 36 tok/s | Not rated |
| Gemma 3 4B IT | 8-bit | 128K | 24 tok/s | Text and images |
| Llama 3.2 3B Instruct | 8-bit | 56K | 27 tok/s | Not rated |
| LightOnOCR 2 1B | 8-bit | 16K | 76 tok/s | Text and images |
| Qwen3 1.7B Base | 8-bit | 32K | 48 tok/s | Not rated |
| HunyuanOCR | 8-bit | 88K | 71 tok/s | Text and images |
| Qwen3 0.6B Base | 8-bit | 32K | 115 tok/s | Not rated |
| Rax 4.5 | 8-bit | 256K | 37 tok/s | Text and images |
| Qwen2 VL 7B Instruct | 8-bit | 20K | 12 tok/s | Text and images |
| Ornith 1.5 9B | 6-bit | 56K | 13 tok/s | Not rated |
| Unlimited OCR 2 | 8-bit | 104K | 71 tok/s | Reading documents |
| Granite 4.1 3B | 8-bit | 72K | 25 tok/s | Not rated |
| DeepSeek R1 Distill Qwen 1.5B | 8-bit | 128K | 53 tok/s | Not rated |
| Mistral 7B V0.1 | 8-bit | 32K | 13 tok/s | Not rated |
| Qwen2.5 Coder 3B Instruct | 8-bit | 32K | 28 tok/s | Coding |
| Meta Llama 3 8B Instruct Abliterated V3 | 8-bit | 8K | 12 tok/s | Not rated |
| EXAONE 3.5 7.8B Instruct | 6-bit | 28K | 16 tok/s | Not rated |
| Saiga Llama3 8B | 8-bit | 8K | 12 tok/s | Not rated |
| MiniCPM V 4.7 1B | 8-bit | 256K | 62 tok/s | Text and images |
| Qwen2.5 Coder 1.5B Instruct | 8-bit | 32K | 53 tok/s | Coding |
| DeepSeek R1 0528 Qwen3 8B | 6-bit | 20K | 15 tok/s | Not rated |
| Gemma 3 12B IT | 4-bit | 32K | 14 tok/s | Text and images |
| Phi 3.5 Mini Instruct | 6-bit | 16K | 30 tok/s | Not rated |
| Qwen3 4B Thinking 2507 | 8-bit | 36K | 22 tok/s | Not rated |
| Phi 4 Mini Instruct | 8-bit | 44K | 23 tok/s | Not rated |
| Dots.ocr | 8-bit | 128K | 31 tok/s | Reading documents |
| DeepSeek R1 Distill Qwen 7B | 8-bit | 32K | 13 tok/s | Not rated |
| Phi 3 Mini 4K Instruct | 8-bit | 4K | 23 tok/s | Not rated |
| DeepSeek OCR 2 | 8-bit | 8K | 69 tok/s | Reading documents |
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 242 are shown.
Image and video models
| Model | Best fit | Per output | Strongest at |
|---|---|---|---|
| Stable Diffusion XL 1.0 | 16-bit | 2.7 min | Images |
| Z-Image Turbo | 4-bit | 4.5 min | Images |
| FLUX.3 Klein | 4-bit | 6.6 min | Images |
Only just too big
- Gemma 4 26B A4B
- Devstral Small 2 24B
- LFM2.5 24B A2B
- Mistral Small 3.2 24B
- gpt-oss 20B
- Qwen2.5 Coder 14B Instruct
- Qwen2.5 14B Instruct
- Diffusiongemma 26B A4B IT
- GPT Neox 20B
- Phi 4
- DeepSeek R1 Distill Qwen 14B
- Qwen2.5 14B Instruct 1M
- CodeLlama 7B HF
- Qwen1.5 7B
- Deepseek Coder 6.7B Base
- Stable Diffusion 3.5 Large
- Qwen 7B Chat
- Phi 4 Reasoning
- Molmo2 O 7B
- Internlm2 Chat 20B
- Wan 2.2 TI2V-5B
- Internlm2 Chat 20B ExPO
- CodeLlama 13B Instruct HF
These need up to a quarter more memory than the Mac mini M4 16 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 16 GB?
Of the open models we track, the strongest that run well on the Mac mini M4 16 GB are Qwen3.5 9B, Gemma 4 12B, Qwen3.5 4B. 242 text models fit it in all.
Calculated from published specs, not benchmarked. All machines