Skip to main content
Local LLM Guide

Model

Gemma 3 12B Instruct

Start with a supported compressed build and short context. 32GB RAM estimate. Best use: higher-quality local multimodal chat

12BMultimodal

Best use

higher-quality local multimodal chat · image understanding

Memory planning

32GB RAM estimate

Expected pace

Depends on your hardware, app, and conversation length

License

Gemma Terms of Use; read the model’s terms before use

Who should use this model?

  • higher-quality local multimodal chat
  • image understanding
  • long-context work
  • structured output

Choose something else when

  • 8GB RAM or 8GB VRAM is the hard ceiling
  • a text-only runner is required
  • very low latency is required

What hardware makes sense?

The download is only part of the memory needed. Leave room for the app, the input it processes, and other open programs. For chat models, context includes the instructions, conversation, and answer. Longer context needs more memory.

CPU only
Memory or response time may limit use. Try a smaller model first.
8GB system memory
Use a smaller model for this memory limit.
16GB system memory
Memory or response time may limit use. Try a smaller model first.
32GB system memory
A possible starting point with a compressed model and short conversation; check the exact file size.
8GB GPU memory
Use a smaller model for this memory limit.
12GB GPU memory
Memory or response time may limit use. Try a smaller model first.
24GB GPU memory
A possible starting point with a compressed model and short conversation; check the exact file size.
Apple unified memory
A possible starting point with a compressed model and short conversation; check the exact file size.

Model facts that change the decision

Parameters
12B
Input
Text · Image · Video
Published context limit
131,072 tokens
Model format
safetensors · GGUF · Ollama · Transformers · Kaggle
Suggested compression
Q4_K_M for default local use; Q5_K_M if the model and multimodal projection fit comfortably in RAM/VRAM.
Typical Q4 file
7.3GB before runtime and context overhead
Languages
English · 140+ languages

Using image input

This model needs an app or library that supports its image-processing components. Follow the exact model’s instructions for preparing image inputs. Some GGUF downloads also need a matching projector file; loading the text model alone does not establish image support.

First setup

Ollama

ollama run gemma3:12b

LM Studio: Search for bartowski/google_gemma-3-12b-it-GGUF and verify image/mmproj support before enabling vision.

llama.cpp

Use hf.co/bartowski/google_gemma-3-12b-it-GGUF:Q4_K_M plus the matching mmproj file for image input.

Conditions that change the answer

  • Video understanding is described by Google for Gemma 3 generally, but local GGUF/Ollama support should be treated as runner-dependent.
  • GGUF files are community conversions, not official Google GGUF releases.
  • The download size includes the Q4_K_M model plus a matching multimodal projection file where applicable.
  • Allow memory for the model file, the app, and your conversation.

Read the controlling license text →