Gemma
18 models · Google
Google's open-weight small-model family (270M to ~31B, including MoE 'A4B' and efficient E2B/E4B variants, some with reasoning modes) distilled from Gemini tech for local and on-device use.
Models in this family
- Gemma 3 12B InstructGoogle12B parametersOpen weights
- Gemma 3 1B InstructGoogle1B parametersOpen weights
- Gemma 3 270MGoogle0.3B parametersOpen weights
- Gemma 3 27B InstructGoogle27B parametersOpen weights
- Gemma 3 4B InstructGoogle4B parametersOpen weights
- Gemma 3n E2B InstructGoogleOpen weights
- Gemma 3n E4B InstructGoogleOpen weights
- Gemma 3n E4B Instruct Preview (May '25)GoogleOpen weights
- Gemma 4 12B (Non-reasoning)Google12B parametersOpen weights
- Gemma 4 12B (Reasoning)Google12B parametersOpen weights
- Gemma 4 26B A4B (Non-reasoning)Google26B total / 4B active (MoE)Open weights
- Gemma 4 26B A4B (Reasoning)Google26B total / 4B active (MoE)Open weights
- Gemma 4 31B (Non-reasoning)Google31B parametersOpen weights
- Gemma 4 31B (Reasoning)Google31B parametersOpen weights
- Gemma 4 E2B (Non-reasoning)GoogleOpen weights
- Gemma 4 E2B (Reasoning)GoogleOpen weights
- Gemma 4 E4B (Non-reasoning)GoogleOpen weights
- Gemma 4 E4B (Reasoning)GoogleOpen weights