Google Unveils Gemma 4, Its Most Capable Open AI Models
Google releases Gemma 4, a highly efficient family of open models built for advanced reasoning and agentic workflows. The new models deliver state-of-the-art performance while requiring significantly less computing power.
Google introduces Gemma 4, marking its most intelligent open models to date. Purpose-built for advanced reasoning and agentic workflows, this new release delivers an unprecedented level of intelligence per parameter. The launch builds on massive community momentum, as developers have downloaded Gemma models over 400 million times to create more than 100,000 unique variants.
The Gemma 4 family arrives in four versatile sizes: Effective 2B, Effective 4B, 26B Mixture of Experts, and 31B Dense. The larger models achieve state-of-the-art rankings on the Arena AI text leaderboard, with the 31B model claiming the number three spot globally among open models. Remarkably, these models outcompete alternatives that are twenty times their size, allowing developers to achieve frontier-level capabilities with much less hardware overhead.
Built from the same foundational research as Gemini 3, Gemma 4 operates under an open Apache 2.0 license to maximize accessibility. The smaller E2B and E4B models focus specifically on edge computing, prioritizing multimodal capabilities, low-latency processing, and seamless ecosystem integration for mobile devices. Together, these open models complement Google's proprietary Gemini lineup, giving developers a powerful combination of tools.