Gemma 4 Released Under Apache 2.0 — Google's Most Capable Open Models in Four Sizes

Google Blog·high signal

Google released Gemma 4, open models built from the same technology as Gemini 3, in four sizes: E2B, E4B, 26B MoE, and 31B Dense. All support images/video input; small variants also support audio. Features include 256K context, 140+ language support, and purpose-built agentic/reasoning workflows. The 31B model ranks #3 on Arena AI text leaderboard among open models. Available on Kaggle, Hugging Face, and Ollama under Apache 2.0.