Find the Best Local AI Model for Your Device
Tell us your hardware and get the single best local LLM it can actually run: one verified pick sized to your exact memory, not a generic chatbot guess. Honest speed and quality estimates for Apple Silicon, iPhone, NVIDIA, and AMD.
Or jump to any device, try: RTX 4090, MacBook Pro M5, 16GB, iPhone 17 Pro
Install the optimized LLM for your machine in one line
ModelFit's free, zero-dependency CLI detects your exact hardware, picks the single best local model it can run, and hands you the one-line Ollama command to install it. The dataset behind it is open (CC BY 4.0), and AI agents can query it live over MCP.
Latest from the blog
How much LLM fits your memory?
A local LLM needs roughly 0.6 GB per billion parameters at Q4. ModelFit sizes picks to ~70% of your unified memory on machines up to 32GB, scaling to ~85% at 128GB+, leaving room for the OS, context, and KV-cache. Here is what each tier comfortably runs today. The same math applies to GPU VRAM: for dedicated cards, see the per-card GPU guides.
| Unified memory | Runs up to (dense) | Top local pick | Models that fit |
|---|---|---|---|
| 8 GB | ~9B params | Ornith 1.0 9B | 32 of 106 |
| 12 GB | ~12B params | Gemma 4 12B | 44 of 106 |
| 16 GB | ~14B params | Qwen3.5 9B Instruct (Q8) | 52 of 106 |
| 24 GB | ~29.3B params | Qwen3.8 27B | 65 of 106 |
| 32 GB | ~35B params | Qwen3.6 35B-A3B | 77 of 106 |
| 36 GB | ~35B params | Qwen3.6 35B-A3B | 81 of 106 |
| 48 GB | ~35B params / 46.7B MoE | Qwen3.8 27B (Q8) | 88 of 106 |
| 64 GB | ~70B params | Qwen3.6 35B-A3B (Q8) | 93 of 106 |
| 72 GB | ~70B params / 80B MoE | Qwen3.6 35B-A3B (Q8) | 94 of 106 |
| 96 GB | ~70B params / 122B MoE | Qwen3.5 122B-A10B Instruct | 99 of 106 |
| 128 GB | ~70B params / 122B MoE | Qwen3.5 122B-A10B Instruct | 101 of 106 |
| 192 GB | ~70B params / 235B MoE | Qwen3 235B A22B | 103 of 106 |
| 256 GB | ~70B params / 235B MoE | Qwen3 235B A22B | 103 of 106 |
| 512 GB | ~405B params / 671B MoE | Qwen3 235B A22B | 106 of 106 |
Derived from ModelFit's own catalog of 143 models · dataset updated 2026-09-18 · full dataset (CC BY 4.0)
Run your own number: VRAM calculator, per-GPU answers, VRAM per model size, RAM vs VRAM or the full memory calculator.
Local vs Cloud: How Close Are We?
Open-weight models you can run at home keep narrowing the gap with the best closed APIs on coding. Here is where the line sits today on SWE-Bench Verified.
SWE-Bench Verified · third-party scores, each raw-confirmed against the model's primary source. Qwen3.6-27B runs locally on a 24GB Mac.
Browse the full index
Local LLM questions, answered
How much memory do I need to run a local LLM?
A local model needs roughly 0.6 GB of memory per billion parameters at Q4 quantization for the weights, closer to 0.8 GB per billion under 8B once runtime overhead is counted. ModelFit sizes its picks to about 70% of your device's unified memory (scaling to ~85% on 128GB+ machines), so an 8GB machine comfortably runs dense models up to ~9B and a 16GB machine up to ~14B parameters. Mixture-of-experts models fit larger total parameter counts because only a fraction is active per token.
What is the best local LLM for a 16GB Mac?
On a 16GB Mac, ModelFit's top pick is Qwen3.5 9B Instruct (Q8): it loads in about 10.7GB and is the highest-quality local model that fits comfortably. 52 of ModelFit's 106 local models fit a 16GB device.
Can I run these models completely offline?
Yes. Every model ModelFit recommends runs fully offline through Ollama on your own hardware: no API keys, no subscription, and no data leaves your machine.
How many AI models does ModelFit cover?
ModelFit tracks 143 AI models across 24 families; 106 run locally via Ollama on Apple Silicon, NVIDIA, or AMD hardware.
How accurate are the speed estimates?
Tokens-per-second figures are ModelFit estimates derived from your chip or GPU's memory bandwidth and the model's size, not measured benchmarks. Real-world speed varies with quantization, context length, and thermal state.