Google Releases Gemma 4 Under Apache 2.0 License

Google launched Gemma 4, a family of four open-weight models ranging from 2B and 4B edge variants to 26B mixture-of-experts and 31B dense models, all under Apache 2.0 - a significant licensing shift from previous Gemma releases. The 31B and 26B models claimed third and sixth spots on Arena AI's text leaderboard, beating models 20 times their size. All variants handle video and images, while the smaller edge models also process audio and speech. Context windows go up to 256K for the larger models. The Apache 2.0 move matters more than the benchmarks for most teams - it removes the commercial restrictions that made earlier Gemma versions awkward for production use. Meta's Llama, Mistral, and now Google are all converging on permissive licensing as the default for open-weight releases.