# Google Releases Gemma 4 Under Apache 2.0, Its Most Capable Open-Source AI Models Yet

*2026-04-07 01:02 UTC · Technology · 1 min read*

> Google DeepMind launched Gemma 4 with four model sizes up to 31B parameters, multimodal capabilities, and fully permissive Apache 2.0 licensing.

Google DeepMind released Gemma 4 on April 2, a family of four open-source models ranging from a compact 5.1B-parameter edge variant to a 30.7B-parameter dense model. The biggest shift from previous Gemma generations is licensing: all Gemma 4 models ship under a fully permissive Apache 2.0 license, removing all restrictions on commercial use, modification, and redistribution.

The 31B dense model ranks #3 globally on the Arena AI text leaderboard and delivers a dramatic leap in math reasoning, jumping from 20.8% to 89.2% on the AIME 2026 benchmark. All four models natively process text and images, while the smaller E2B and E4B variants add audio input for speech recognition. Video processing up to 60 seconds, function calling for autonomous agents, and support for over 140 languages round out the feature set, with context windows stretching to 256K tokens on the larger models.

The release intensifies the open-source AI race against Meta's Llama and Alibaba's Qwen families, and the Apache 2.0 move removes a key complaint developers had about earlier Gemma licenses. Models are available now on Hugging Face, Kaggle, Ollama, and Google AI Studio.

## Sources

- [blog.google](https://blog.google/innovation-and-ai/technology/developers-tools/gemma-4/)
- [cometapi.com](https://www.cometapi.com/google-releases-gemma-4-open-source-model/)
- [news.cgtn.com](https://news.cgtn.com/news/2026-04-03/Google-launches-open-model-Gemma-4-1M2WodjFYw8/p.html)

---
TechTools News · https://techtools.cz/tools/news/article/1323-google-releases-gemma-4-under-apache-2-0-its-most-capable-open-source-ai-models
Česká verze: https://techtools.cz/tools/news/clanek/1378-google-vydava-gemma-4-pod-licenci-apache-2-0-sve-dosud-nejschopnejsi-open
