Macs
Which AI models run on the M3 (8 GB)?
The M3 (8 GB) has 8 GB of memory, about 5.3 GB of it usable for a model, and 100 GB/s of memory bandwidth. 20 of the 110 models in our catalog run on it at 4-bit or better.
Data as of 2 November 2026 · Vendor specs
Models that run on it
| Model | Best build | Memory | Speed |
|---|---|---|---|
| Gemma 3 4B Instruct4B | Q8_0 | 5.0 GB | ~20 tok/s ModelFit est. |
| Gemma 4 E4B4.5B | Q4_K_M | 5.1 GB | ~24 tok/s ModelFit est. |
| Qwen3.5 4B Instruct4B | Q4_K_M | 4.6 GB | ~30 tok/s ModelFit est. |
| Phi-3 Mini 3.8B3.8B | Q4_K_M | 4.3 GB | ~17 tok/s our est. |
| Phi-4 Mini 3.8B3.8B | Q8_0 | 4.9 GB | ~18 tok/s ModelFit est. |
| Granite 4.2 3B3.7B | Q4_K_M | 3.2 GB | ~34 tok/s ModelFit est. |
| Ministral 3 3B3.4B | Q4_K_M | 4.1 GB | ~25 tok/s ModelFit est. |
| Qwen2.5 3B Instruct3B | Q4_K_M | 3.6 GB | ~40 tok/s ModelFit est. |
| Llama 3.2 3B Instruct3B | Q4_K_M | 3.6 GB | ~33 tok/s ModelFit est. |
| Granite 4.1 3B Instruct3B | Q4_K_M | 3.0 GB | ~34 tok/s ModelFit est. |
| Gemma 4 E2B2.3B | Q4_K_M | 3.4 GB | ~55 tok/s ModelFit est. |
| Gemma 2 2B Instruct2B | Q4_K_M | 2.8 GB | ~38 tok/s ModelFit est. |
| Qwen3.5 2B Instruct2B | Q4_K_M | 2.8 GB | ~40 tok/s ModelFit est. |
| LFM3 1.7B1.73B | Q4_K_M | 2.1 GB | ~69 tok/s ModelFit est. |
| Qwen2.5 1.5B Instruct1.5B | Q4_K_M | 2.5 GB | ~72 tok/s ModelFit est. |
| Llama 3.2 1B Instruct1B | Q4_K_M | 2.0 GB | ~82 tok/s ModelFit est. |
| Gemma 3 1B Instruct1B | Q8_0 | 2.0 GB | ~72 tok/s ModelFit est. |
| Qwen3.5 0.8B Instruct0.8B | Q4_K_M | 1.8 GB | ~90 tok/s ModelFit est. |
| Qwen2.5 0.5B Instruct0.5B | Q4_K_M | 1.8 GB | ~156 tok/s ModelFit est. |
| SmolLM2 360M0.36B | Q4_K_M | 1.5 GB | ~86 tok/s ModelFit est. |
Speeds come from ModelFit's estimates (CC BY 4.0) where it covers the model, otherwise from our bandwidth formula. Method
Fits only at 3-bit or lower
Too big for this machine
Mistral 7B Instruct (5.9 GB), Granite 4.2 8B (6.1 GB), LFM2.5 8B-A1B (6.6 GB), Granite 4.1 8B Instruct (6.6 GB), Qwen2.5 7B Instruct (6.6 GB), Qwen2.5 Coder 7B (6.6 GB), DeepSeek-R1 Distill Qwen 7B (6.6 GB), Ornith 1.0 9B (6.7 GB), MiniCPM-V 4.5 8B (6.8 GB), MiMo-V2.6-Distill-Qwen-9B (6.9 GB), Jebadiah 9B (6.9 GB), Ministral 3 8B (7.1 GB) and more.
Questions
What is the best AI model for the M3 (8 GB)?
Gemma 3 4B Instruct at Q8_0: the largest model on our shortlist of current models that runs at 4-bit or better at a usable speed (about 20 tok/s, ModelFit est.).
How much memory can a model use on the M3 (8 GB)?
About 5.3 GB of the 8 GB.
How fast is the M3 (8 GB) for local LLMs?
Decode speed is limited by memory bandwidth: 100 GB/s. A model that reads 10 GB per token runs at under 10 tokens per second.