With 16 GB unified memory (about 10.7 GB usable by the GPU), these are the best open models that fit. The top pick is Mistral Nemo 12B at Q5_K_M (~9.9 GB, ~9 tok/s): ollama run mistral-nemo:12b-instruct-2407-q5_K_M.
Best for coding: DeepSeek Coder V2 Lite 16B (MoE) (deepseek-coder-v2:16b), Qwen 2.5 Coder 14B (qwen2.5-coder:14b), Qwen 2.5 Coder 7B (qwen2.5-coder:7b-instruct-q8_0).
Runtimes available for this hardware: Ollama, LM Studio, llama.cpp, MLX.
Open the interactive checker for M2 16GB| # | Model | Quant | Memory needed | Fit | Speed | Command |
|---|---|---|---|---|---|---|
| 1 | Mistral Nemo 12B | Q5_K_M | 9.9 GB | fits in fast memory | ~9 tok/s | ollama run mistral-nemo:12b-instruct-2407-q5_K_M |
| 2 | DeepSeek R1 Distill Qwen 14B | Q4_K_M | 10.3 GB | fits in fast memory | ~9 tok/s | ollama run deepseek-r1:14b |
| 3 | Phi-4 14B | Q4_K_M | 10.4 GB | fits in fast memory | ~9 tok/s | ollama run phi4:14b |
| 4 | Gemma 3 12B | Q4_K_M | 9.4 GB | fits in fast memory | ~10 tok/s | ollama run gemma3:12b |
| 5 | Qwen3 14B | Q4_K_M | 10.6 GB | fits in fast memory | ~9 tok/s | ollama run qwen3:14b |
| 6 | Llama 3.2 Vision 11B | Q4_K_M | 8.9 GB | fits in fast memory | ~10 tok/s | ollama run llama3.2-vision:11b |
| 7 | DeepSeek R1 Distill Qwen 7B | Q8_0 | 9.1 GB | fits in fast memory | ~10 tok/s | ollama run deepseek-r1:7b-qwen-distill-q8_0 |
| 8 | DeepSeek Coder V2 Lite 16B (MoE) | Q4_0 | 9.9 GB | fits in fast memory | ~29 tok/s | ollama run deepseek-coder-v2:16b |
| 9 | Qwen3 8B | Q8_0 | 9.9 GB | fits in fast memory | ~9 tok/s | ollama run qwen3:8b-q8_0 |
| 10 | DeepSeek R1 0528 Qwen3 8B | Q8_0 | 9.9 GB | fits in fast memory | ~9 tok/s | ollama run deepseek-r1:8b-0528-qwen3-q8_0 |
| 11 | Nimble v3 9B (Bespoke Labs, decision) | Q8_0 | 10.6 GB | fits in fast memory | ~8 tok/s | ollama run nimble-v3:9b-q8_0 |
| 12 | MiniCPM-V 2.6 8B | Q8_0 | 10.1 GB | fits in fast memory | ~9 tok/s | ollama run minicpm-v:8b-2.6-q8_0 |
Estimates, not benchmarks. Model sizes are ollama.com download sizes; memory needed adds headroom for ~8k tokens of context; speeds are rough figures from memory bandwidth. Data updated 2026-10-14.
A project by OnDevice Labs · llms.txt