Best local AI models for MacBook Air M4 (24 GB)
A fanless laptop that runs 7–14B models. Fine for chat and retrieval, not for training.
Enough memory to hold an adapter run, but too little bandwidth to finish one in reasonable time.
Specification
Unified memory24 GB
Usable for models18 GB
System RAM24 GB
Bandwidth120 GB/s
Local trainingNot practical
Operating systemsmacos
Runs well
27 models fit this machine
Estimated
| Model | Params | Quantization | Memory | ~tok/s | Fit | Fine-tune |
|---|---|---|---|---|---|---|
| Phi-4 14B | 14.7B | Q5_K_M | 12 GB | 9 | Good fit | Yes |
| Qwen2.5 14B Instruct | 14.8B | Q5_K_M | 12 GB | 9 | Good fit | Yes |
| Mistral Nemo 12B Instruct | 12.2B | Q4_K_M | 9.1 GB | 13 | Excellent fit | Yes |
| Qwen2.5 Coder 7B Instruct | 7.6B | Q8_0 | 8.6 GB | 11 | Excellent fit | Yes |
| Qwen3 14B | 14.8B | Q4_K_M | 10 GB | 10 | Excellent fit | Yes |
| StarCoder2 15B | 16B | Q6_K | 14 GB | 7 | Good fit | Yes |
| DeepSeek-R1-Distill-Qwen-14B | 14.8B | Q5_K_M | 12 GB | 9 | Good fit | Yes |
| Qwen2.5 7B Instruct | 7.6B | Q8_0 | 8.6 GB | 11 | Excellent fit | Yes |
| Llama 3.1 8B Instruct | 8B | Q8_0 | 9.5 GB | 11 | Excellent fit | Yes |
| Gemma 3 12B Instruct | 12.2B | Q5_K_M | 12 GB | 11 | Good fit | Yes |
| DeepSeek-Coder-V2-Lite Instruct | 15.7B | Q4_K_M | 11 GB | 64 | Good fit | Yes |
| Gemma 2 9B Instruct | 9.2B | Q8_0 | 12 GB | 9 | Good fit | Yes |
| Qwen3 8B | 8.2B | Q8_0 | 9.8 GB | 11 | Excellent fit | Yes |
| Mistral 7B Instruct v0.3 | 7.25B | Q8_0 | 8.8 GB | 12 | Excellent fit | Yes |
| Code Llama 7B Instruct | 6.7B | Q6_K | 9.7 GB | 17 | Excellent fit | Yes |
| Mistral Small 24B Instruct | 23.6B | Q4_K_M | 16 GB | 7 | Tight fit | Yes |
| Qwen3 30B-A3B | 30.5B | Q3_K_M | 15 GB | 58 | Tight fit | No |
| Phi-3.5 Mini Instruct | 3.8B | F16 | 11 GB | 12 | Excellent fit | Yes |
Fine-tunable here
Models you can train on this machine
Out of reach