Google Gemma

Google's open-weight model family for developers who need to self-host or fine-tune. Distilled from Gemini research, Gemma 4 ships in four sizes (E2B to 31B Dense), uses a hybrid local/global attention architecture with up to 256K context, and the 31B model ranks #3 on Arena AI's open-model leaderboard under a fully commercial Apache 2.0 license.