Guides · Updated 2026-10-17
Best local AI models for a 12GB GPU (2026)
What runs on an RTX 3060 12GB, RTX 5070 or Arc B580: the best 9B-14B Ollama models by use case.
Not sure which model to pull? Use the GPU model picker: enter your VRAM and RAM and get a list of Ollama models that fit.
12 GB is the sweet spot for budget local AI. You can run 12-14B models fully on the GPU, which are noticeably smarter than 7-8B ones, or 8-9B models with long context.
Top picks
| Use | Model | Command | Download |
|---|---|---|---|
| All-rounder | Gemma 4 12B | ollama run gemma4:12b | 8.0 GB |
| All-rounder, long context | Qwen3.5 9B | ollama run qwen3.5:9b | 6.6 GB |
| Writing & documents | Ministral 3 14B | ollama run ministral-3:14b | 9.1 GB |
| Reasoning | Qwen3 14B | ollama run qwen3:14b | 9.3 GB |
| Maths & logic | Phi-4 14B | ollama run phi4 | 9.1 GB |
| Coding | Qwen2.5-Coder 14B | ollama run qwen2.5-coder:14b | 9.0 GB |
| Vision | Gemma 4 12B or Qwen3-VL 8B | ollama run qwen3-vl:8b | 6.1 GB |
Tips
- 9-10 GB models leave ~2 GB for context. Fine for chat; for long documents step down to Qwen3.5 9B.
- The RTX 5070 is roughly twice as fast as an RTX 3060 on the same model, because its memory bandwidth is 672 GB/s vs 360 GB/s. Both run the same models.
- Coding: pair a chat model with
qwen2.5-coder:1.5bfor tab autocomplete. See the coding assistant guide.
Check your setup in the GPU picker, or see 12GB hardware picks.
Get the Local AI weekly
New models worth running, what fits which GPU, and setup fixes. One short email a week, free. Unsubscribe any time.
Local-LLM App Starter
A tested starter for building your own app on Ollama: chat, RAG and a GPU-fit checker. Skip the setup hours.
kits.lilaforge.com → VibeEngineBest local LLM tools, ranked
Ollama, LM Studio, Open WebUI and llama.cpp scored from public evidence, with top picks.
vibeengine.lilaforge.com →