Can I run an LLM on my Mac?
Short answer: yes, on Apple Silicon. The GPU uses unified memory, and macOS lets it use about 67% below 64 GB and about 75% from 64 GB up. A 16GB Mac fits Phi 4 reasoning vision 15B (est.); a 128GB Mac fits Qwen3.8 Flash Next (MoE) (est.).
Checked 2026-10-22 · memory figures are estimates
Pick your memory
| Mac memory | GPU can use (approx.) | Models that fit (8k ctx, est.) | Largest that fits |
|---|---|---|---|
| 16 GB | ~11 GB | 17 | Phi 4 reasoning vision 15B (IQ4_XS) |
| 24 GB | ~16 GB | 25 | Raon-1 mini preview 28.8B A4.3B (MoE) (Q3_K_M) |
| 32 GB | ~21 GB | 35 | Qwen3.6 35B A3B (MoE) (IQ4_XS) |
| 48 GB | ~32 GB | 37 | Kimi Linear 48B A3B (MoE) (Q4_K_M) |
| 64 GB | ~48 GB | 39 | Qwen3 Coder Next (MoE) (IQ4_XS) |
| 96 GB | ~72 GB | 47 | Mistral Medium 3.5 128B (Q3_K_M) |
| 128 GB | ~96 GB | 48 | Qwen3.8 Flash Next (MoE) (Q3_K_M) |
Common questions
- Is 16GB RAM enough to run an LLM locally?
- MLX vs GGUF on a Mac: which is faster?
- How much of a Mac's memory can the GPU use?
Model on Mac pages
gpt-oss 20B (MoE) on Apple M4 Pro (48 GB) · gpt-oss 20B (MoE) on Apple M4 Max (128 GB) · gpt oss 120b (MoE) on Apple M4 Max (128 GB) · Qwen3.6 27B on Apple M4 Pro (48 GB) · Qwen3.6 27B on Apple M4 Max (128 GB) · Qwen3.6 35B A3B (MoE) on Apple M4 Pro (48 GB) · Qwen3.6 35B A3B (MoE) on Apple M4 Max (128 GB) · Qwen3.8 27B on Apple M4 Pro (48 GB) · Qwen3.8 27B on Apple M4 Max (128 GB) · Gemma 4 31B on Apple M4 Pro (48 GB) · Gemma 4 31B on Apple M4 Max (128 GB) · Gemma 4 26B A4B (MoE) on Apple M4 Pro (48 GB) · Gemma 4 26B A4B (MoE) on Apple M4 Max (128 GB) · Gemma 4 12B on Apple M4 Pro (48 GB) · Qwen3.5 9B on Apple M4 Pro (48 GB) · Qwen3 Coder 30B A3B (MoE) on Apple M4 Pro (48 GB) · Qwen3 Coder 30B A3B (MoE) on Apple M4 Max (128 GB) · GLM 4.7 Flash on Apple M4 Pro (48 GB) · GLM 4.7 Flash on Apple M4 Max (128 GB) · Llama 3.1 8B Instruct on Apple M4 Pro (48 GB) · Llama 3.1 8B Instruct on Apple M4 Max (128 GB) · Qwen3.5 9B on Apple M4 (16 GB) · Gemma 4 12B on Apple M4 (16 GB) · Llama 3.1 8B Instruct on Apple M4 (16 GB) · Qwen3 VL 8B on Apple M4 (16 GB) · gpt-oss 20B (MoE) on Apple M4 Pro (24 GB) · Gemma 4 12B on Apple M4 Pro (24 GB) · Qwen3.6 27B on Apple M4 (32 GB) · Qwen3 Coder 30B A3B (MoE) on Apple M4 (32 GB) · GLM 4.7 Flash on Apple M4 (32 GB) · Devstral Small 2 24B (2512) on Apple M4 (32 GB) · Olmo 3.1 32B on Apple M4 Pro (48 GB) · Kimi Linear 48B A3B (MoE) on Apple M4 Pro (48 GB) · NVIDIA Nemotron 3 Nano 30B A3B (MoE) on Apple M4 Pro (48 GB) · Qwen3.6 35B A3B (MoE) on Apple M4 Max (64 GB) · GLM 4.5 Air (MoE) on Apple M3 Ultra (96 GB) · Llama 4 Scout 17B 16E (MoE) on Apple M3 Ultra (96 GB) · Qwen3 Coder Next (MoE) on Apple M3 Ultra (96 GB) · Llama 3.3 70B Instruct on Apple M3 Ultra (96 GB) · Llama 3.3 70B Instruct on Apple M4 Max (128 GB) · GLM 4.5 Air (MoE) on Apple M4 Max (128 GB) · Mistral Medium 3.5 128B on Apple M4 Max (128 GB) · NVIDIA Nemotron 3 Super 120B A12B (MoE) on Apple M4 Max (128 GB) · Ornith 1.5 122B (MoE) on Apple M4 Max (128 GB)
What fits my GPU or Mac? · What to buy
Will It Run? Pro NOT BUILT YET
A personal setup for your exact hardware (Ollama / llama.cpp configs ready for your coding agent), an alert when a new model fits, and a one-click cloud fallback when it doesn't.
Not built yet. Join the waitlist, $10/mo planned, no charge now. What's planned
Email only. No payment info.