Gemma 4 (E2B, E4B, 26B MoE, 31B)
Gemma 4 ships under Apache 2.0 in four sizes; the 31B dense ranked #3 among open models on Arena AI text, the 26B MoE #6.
Built from Gemini 3 research for reasoning and agent workflows, with 128K context on edge models and up to 256K on larger ones, 140+ languages, and native audio input on E2B and E4B. Apache 2.0 replaces the custom Gemma terms of use.
- Date
- Thursday, 2 April 2026
- Lab
- Google DeepMind
- Kind
- open-weights
- Access
- open weights
Figures
| Measure | Value | Measured by |
|---|---|---|
| Arena AI text leaderboard (open models) | #3 (31B), #6 (26B) as of 2026-04-01; Google says it beats models 20x larger | Arena AI |
License moves from the custom Gemma terms of use (Gemma 1-3) to Apache 2.0. Gemma 4 12B (2026-06-03) and DiffusionGemma (2026-06-10) followed.
Sources
- deepmind.google/blog/gemma-4-byte-for-byte-the-most-capable-open-models/
- ai.google.dev/gemini-api/docs/changelog
- blog.google/innovation-and-ai/technology/ai/google-ai-updates-april-2026/
This record was checked against its sources on 6 October 2026. How we check