§01·spec · /gpus
Apple M2 Pro
appleapple series16GB unified
13 open-weights AI models run on the Apple M2 Pro — see which fit, how fast they go, and the VRAM each needs.
Apple Silicon uses unified memory shared between CPU and GPU. By default the GPU can address about 75% of it; the limit is raisable via Metal’s wired-memory setting.
§02·models that run on this GPU
13 totalLLM · 8
| Model | Best speed | Min VRAM | Works | Evidence | |
|---|---|---|---|---|---|
| Gemma 4 12B | 16GB | ✓ | recipe | check ↗ | |
| gpt-oss 20B | 13GB | ✓ | recipe | check ↗ | |
| Llama 3.1 8B | 5GB | ✓ | recipe | check ↗ | |
| Mistral Nemo 12B | 16GB | ✓ | recipe | check ↗ | |
| Ornith 1.0 9B | 12GB | ✓ | recipe | check ↗ | |
| Phi-4 | 16GB | ✓ | recipe | check ↗ | |
| Qwen3 14B | 9GB | ✓ | recipe | check ↗ | |
| Qwen3-8B | 5GB | ✓ | recipe | check ↗ |
Multimodal · 1
| Model | Best speed | Min VRAM | Works | Evidence | |
|---|---|---|---|---|---|
| Gemma 4 E4B-IT | 5GB | ✓ | recipe | check ↗ |
Image · 1
| Model | Best speed | Min VRAM | Works | Evidence | |
|---|---|---|---|---|---|
| Z-Image Turbo | 6GB | ✓ | recipe | check ↗ |
§03·tested recipes
showing 6 of 13- llmintermediate16GB+recipe
Gemma 4 12B on Apple M2 Pro: Local Private Assistant via llama.cpp / Ollama (16GB)
- llmintermediate16GB+recipe
Phi-4 (14B) on Apple M2 Pro: Local Private Assistant via llama.cpp / Ollama (16GB)
- llmintermediate16GB+recipe
Mistral Nemo 12B on Apple M2 Pro (16GB): Local Private Assistant via llama.cpp / Ollama (Metal)
- llmintermediate12GB+recipe
Ornith 1.0 9B on Apple M2 Pro: Local Agentic Coding in 16GB Unified Memory via llama.cpp Metal + OpenHands
- videoadvanced14GB+recipe
LTX-2.3 on Apple M2 Pro (16 GB): experimental 22B audio-video in unified memory via the MLX --low-ram port
- ttsbeginner2GB+recipe
VoxCPM-0.5B on Apple M2 Pro: Zero-Shot Voice Cloning TTS in Unified Memory (MPS)