G
Gemma 3
40
Stack Score
Best for
Best for builders needing small, capable, multilingual OSS models.
At a glance
- Pricing
- open-source
- Setup time
- 30 minutes
- Learning curve
- easy
- Last verified
- 5/26/2026
What we love
- ✓Best small-OSS-model family
- ✓Multimodal at every size
- ✓Strong Google support
Where it falls short
- −Gemma license has some commercial constraints
- −Behind Llama in ecosystem
- −Largest size still trails frontier
Pricing
2 tiers
Open-source
$0
perpetual
- ✓Gemma license
Hosted
Usage
per token
- ✓AI Studio
- ✓Vertex
Overview
Gemma 3 is Google DeepMind's open-weights LLM family — 1B, 4B, 12B, 27B variants with 128K context, multimodal (image input), and 140+ languages. Built on Gemini research.
Key features
- ●1B / 4B / 12B / 27B variants
- ●128K context
- ●140+ languages
- ●Multimodal (image input)
- ●Permissive Gemma license
Best use cases
Edge deployment
Run Gemma 4B on a laptop or phone with surprisingly good quality.
Multilingual SaaS
Power multi-language features without per-token API cost.
Integrations
Similar tools
See all LLMs →One AI Stack pick every week.
The tool we’d pay for this week, why we’d pay for it, and what we’d skip. No promo spam.