According to the Gemma 4 model card published on June 10, 2026, the Gemma 4 family includes new model sizes such as E2B, E4B, 26B A4B, and 31B, which are multimodal open models supporting text and image input, with some variants also supporting audio input.
Open source ↗
Google
Gemma 次期モデル
PREV Gemma 4 12B Unified · 2026-06-03
- Window
- Nov 2026 – Feb 2027
- Confidence
- 35%
- Analysis
- Gemma has progressed from the February 2024 initial release to Gemma 2 in June 2024, Gemma 3 in March 2025, Gemma 4 in March 2026, and the Gemma 4 12B Unified addition in June 2026. Gemma 4 31B is treated as an already released history model, so the next official name remains unset while we keep a low-confidence watch for the next official Gemma family.
Last updated: 2026-06-24
Gemma family release history
Historical release interval
2–315d
Est. next interval
9d×0.64/release
Days since current model
22d
Current model released
2026-06-03
Signals
Google DeepMind announced DiffusionGemma, an experimental open model that uses text diffusion methods and achieves up to 4x faster inference compared to typical local LLMs when using a GPU.
Open source ↗›Prediction history (2)
2026-06-24 · Jul 1 – Jul 31, 2026 · 50%
The forecast has been pulled forward significantly based on an official Google model card. This document explicitly lists several unreleased multimodal models in the Gemma 4 family, including 26B and 31B variants. The public documentation of these models strongly suggests their release is imminent, justifying the revised July window and increased confidence.
2026-06-24 · Nov 2026 – Feb 2027 · 35%
The forecast has been pulled forward significantly and confidence increased due to an official Google model card. This document explicitly lists several unreleased multimodal models within the Gemma 4 family (including 26B and 31B variants), suggesting their release is being prepared and is likely to occur in the near future.