← Every modelGoogle DeepMind · 12B, dense

Gemma 3 12B

Google's mid-size open model, and the smallest one here that reads images. Strong at writing and at languages other than English.

ollama run gemma3:12b

Needs about 10 GB of graphics memory, or unified memory on a Mac. First time? Install Ollama, then paste that into a terminal.

Download8.1 GB
Memory to runAbout 10 GB
Read per token8.1 GB
LicenseGemma Terms of Use
Context128K tokens
ReleasedMarch 2025
Good for
  • Describing and asking questions about images and screenshots
  • Writing and translation in many languages
  • Long documents, thanks to the 128K context
Not the best pick for
  • Hard reasoning and math, where Qwen3 14B does better
  • Projects that need a standard open-source license: Gemma has its own terms
111 GPUs and Macs

What runs Gemma 3 12B

Estimated speed on each card at the smallest memory size that fits it, from memory bandwidth. As a dense model it reads all its weights for every token, so memory bandwidth sets the speed.

Runs great 73 cards

Usable 30 cards

Slow 8 cards

Cheapest build from our list that runs it well: Used desktop + RTX 3060 12 GB, from $600 used, at about 24 tok/s. See the builds.

Learn

Read more about Gemma 3 12B

Compare

Other models