Gemma 3 (4B / 12B / 27B)

Previous-generation Gemma line. The 4B is still a useful ultra-compact agent model with native vision. Larger sizes are superseded by Gemma 4.

License: Custom Gemma Terms of Use (not Apache — flag for commercial) · Context: 128K · Released: March 12, 2025

The decision in five lines

The call: Skip for local — for agents
Best for: agents
Runs on: 23 hardware picks fit (cheapest: Intel Arc B580 12 GB · $249)
Watch out: If you have the VRAM for Gemma 4 (16+ GB), use that — Gemma 3 is only the honest pick at 4B or below.
Evidence: Estimated · last verified April 2026

4B: PARAMETERS
DENSE: TYPE
128K: CONTEXT
~3 GB (4B) / ~8 GB (12B) / ~17 GB (27B): VRAM AT Q4

Where we recommend this

Every tier slot in the planner where this model is a top or alternate pick. Pulled live from planner.js — when the planner refreshes, this table stays current.

AGENTS · LOW

Gemma 3 4BCompact Gemma for bounded tool loops; keep iterations tight.

The call

Previous-generation Gemma line. The 4B is still a useful ultra-compact agent model with native vision. Larger sizes are superseded by Gemma 4.
When not to use: If you have the VRAM for Gemma 4 (16+ GB), use that — Gemma 3 is only the honest pick at 4B or below. Also: Gemma Terms of Use propagate to fine-tunes and redistribution, so audit before shipping in a commercial product.

Runner notes

Ollama tags `gemma3:4b` / `gemma3:12b` / `gemma3:27b`. Handles vision natively.

License: Custom Gemma Terms of Use (not Apache — flag for commercial)
Released: March 12, 2025
Maker: Google
Model card: huggingface.co/google/gemma-3-27b-it →

Hardware that fits

Every hardware pick whose memory fits this model at the quant we recommend. Sorted cheapest-first — the top row is your best-value fit. Click through for the full buyer’s guide.

Next step

Find-by-model — see what hardware runs this→