Gemma
Represents the separate 2B and 7B first-generation Gemma checkpoints without assigning one parameter count to the whole release.
FAMILY / GOOGLE DEEPMIND
Gemma
Google’s open-model family for local, edge, and customizable deployment.
- First model
- 2024
- Models
- 9
- Latest dated version
- EmbeddingGemma 2
01 / RELEASE SEQUENCE
Family release sequence
- 2024-02-21GemmaLanguage
- 2024-06-27Gemma 2 27BLanguage · Reasoning
- 2025-03-12Gemma 3 27BLanguage · Reasoning
- 2025-06-26Gemma 3n E4BLanguage · Multimodal
- 2026-04-02Gemma 4 26B MoELanguage · Reasoning
- 2026-04-02Gemma 4 31BLanguage · Reasoning
- 2026-06-03Gemma 4 12BLanguage · Reasoning
- 2026-06-10DiffusionGemma 26BLanguage
- 2026-10-06EmbeddingGemma 2Multimodal
02 / GENERATIONS
What changed
Gemma 2 27B
The 27B open Gemma 2 checkpoint, improving reasoning quality while targeting accessible single-accelerator deployment.
Gemma 3 27B
Added image input, a 128K context window, and support for more than 140 languages.
Gemma 3n E4B
Used MatFormer and Per-Layer Embeddings to bring multimodal understanding to mobile devices.
Gemma 4 26B MoE
The Gemma 4 MoE activating only 3.8B parameters per step for local speed and intelligence density.
Gemma 4 31B
The high-quality dense Gemma 4 flagship for local reasoning, coding, and agent workflows.
Gemma 4 12B
A unified encoder-free multimodal model for laptops, filling the gap between E4B and 26B.
DiffusionGemma 26B
An experimental Gemma 4 text-diffusion model exploring lower latency through parallel block generation.
EmbeddingGemma 2
It extends EmbeddingGemma from text embeddings to unified representations across text, code, images, video, and audio, delivering on-device retrieval in a 740M full multimodal model.