Best overallGood fit
Phi-4 14B
Trained largely on curated synthetic data. Punches far above its size on reasoning and maths; the short context limits what you can do with it.
- Params
- 14.7B
- Quant
- Q5_K_M
- Memory
- 12 GB
Reference
Best models for MacBook Air M4 (24 GB)
A fanless laptop that runs 7–14B models. Fine for chat and retrieval, not for training.
27 models fit18 GB usableEstimated
Top picks
Best on MacBook Air M4 (24 GB)
Runner-upGood fit
Qwen2.5 14B Instruct
The sweet spot for 24 GB cards. Meaningfully stronger reasoning than 7B while still fine-tunable locally with QLoRA.
- Params
- 14.8B
- Quant
- Q5_K_M
- Memory
- 12 GB
ThirdExcellent fit
Mistral Nemo 12B Instruct
A 12B with a 128k context and the Tekken tokenizer, which compresses non-English text far better than Llama’s.
- Params
- 12.2B
- Quant
- Q4_K_M
- Memory
- 9.1 GB
Full results
27 models
Estimated
| # | Model | Params | Licence | Quantization | Memory | ~tok/s | Fit | Fine-tune |
|---|---|---|---|---|---|---|---|---|
| 1 | Phi-4 14B | 14.7B | MIT | Q5_K_M | 12 GB | 9 | Good fit | Yes |
| 2 | Qwen2.5 14B Instruct | 14.8B | Apache 2.0 | Q5_K_M | 12 GB | 9 | Good fit | Yes |
| 3 | Mistral Nemo 12B Instruct | 12.2B | Apache 2.0 | Q4_K_M | 9.1 GB | 13 | Excellent fit | Yes |
| 4 | Qwen2.5 Coder 7B Instruct | 7.6B | Apache 2.0 | Q8_0 | 8.6 GB | 11 | Excellent fit | Yes |
| 5 | Qwen3 14B | 14.8B | Apache 2.0 | Q4_K_M | 10 GB | 10 | Excellent fit | Yes |
| 6 | StarCoder2 15B | 16B | BigCode OpenRAIL-M | Q6_K | 14 GB | 7 | Good fit | Yes |
| 7 | DeepSeek-R1-Distill-Qwen-14B | 14.8B | MIT | Q5_K_M | 12 GB | 9 | Good fit | Yes |
| 8 | Qwen2.5 7B Instruct | 7.6B | Apache 2.0 | Q8_0 | 8.6 GB | 11 | Excellent fit | Yes |
| 9 | Llama 3.1 8B Instruct | 8B | Llama 3.1 Community License | Q8_0 | 9.5 GB | 11 | Excellent fit | Yes |
| 10 | Gemma 3 12B Instruct | 12.2B | Gemma Terms of Use | Q5_K_M | 12 GB | 11 | Good fit | Yes |
| 11 | DeepSeek-Coder-V2-Lite Instruct | 15.7B | DeepSeek License | Q4_K_M | 11 GB | 64 | Good fit | Yes |
| 12 | Gemma 2 9B Instruct | 9.2B | Gemma Terms of Use | Q8_0 | 12 GB | 9 | Good fit | Yes |
| 13 | Qwen3 8B | 8.2B | Apache 2.0 | Q8_0 | 9.8 GB | 11 | Excellent fit | Yes |
| 14 | Mistral 7B Instruct v0.3 | 7.25B | Apache 2.0 | Q8_0 | 8.8 GB | 12 | Excellent fit | Yes |
| 15 | Code Llama 7B Instruct | 6.7B | Llama 2 Community License | Q6_K | 9.7 GB | 17 | Excellent fit | Yes |
| 16 | Mistral Small 24B Instruct | 23.6B | Apache 2.0 | Q4_K_M | 16 GB | 7 | Tight fit | Yes |
| 17 | Qwen3 30B-A3B | 30.5B | Apache 2.0 | Q3_K_M | 15 GB | 58 | Tight fit | No |
| 18 | Phi-3.5 Mini Instruct | 3.8B | MIT | F16 | 11 GB | 12 | Excellent fit | Yes |
| 19 | Gemma 3 4B Instruct | 4.3B | Gemma Terms of Use | F16 | 9.9 GB | 11 | Excellent fit | Yes |
| 20 | Llama 3.2 3B Instruct | 3.2B | Llama 3.2 Community License | F16 | 7.3 GB | 14 | Excellent fit | Yes |
| 21 | SmolLM2 1.7B Instruct | 1.7B | Apache 2.0 | F16 | 5.2 GB | 27 | Excellent fit | Yes |
| 22 | Qwen2.5 Coder 32B Instruct | 32.8B | Apache 2.0 | Q3_K_M | 18 GB | 6 | Runs with CPU offload | No |
| 23 | Qwen2.5 32B Instruct | 32.8B | Apache 2.0 | Q3_K_M | 18 GB | 6 | Runs with CPU offload | No |
| 24 | Llama 3.2 1B Instruct | 1.24B | Llama 3.2 Community License | F16 | 3.0 GB | 37 | Excellent fit | Yes |
| 25 | Mixtral 8x7B Instruct | 46.7B | Apache 2.0 | Q4_K_M | 29 GB | 12 | Runs with CPU offload | No |
| 26 | DeepSeek-R1-Distill-Qwen-32B | 32.8B | MIT | Q3_K_M | 18 GB | 6 | Runs with CPU offload | No |
| 27 | Gemma 3 27B Instruct | 27.4B | Gemma Terms of Use | Q3_K_M | 19 GB | 7 | Runs with CPU offload | No |
How this page is produced
Memory is calculated from each model’s published geometry against this machine’s usable capacity. Where the vendor publishes a memory bandwidth figure, an order-of-magnitude token rate is derived from it and labelled as an estimate.
- · Enough memory to hold an adapter run, but too little bandwidth to finish one in reasonable time.
- · 21 of these can also be fine-tuned on this hardware.
Check your own machine
ModelLM can read your actual GPU, VRAM and RAM and size every model against it.
Detect my hardwareRelated