Gemma 4: Google DeepMind's 'Most Capable' Open-Source AI

Google DeepMind has released Gemma 4, a new family of open-source AI models. These models are designed for advanced reasoning and agentic workflows and are available under a commercially permissive license. Here's the thing: The release includes four model sizes, ranging from Effective 2B to a 31B Dense model. All are built on the same research as Google's proprietary Gemini 3 models. The company states Gemma 4 is their "most capable" open-source AI. What's interesting is that developers have already downloaded models in the Gemma family over 400 million times. Google claims the new Gemma 4 models can outcompete rivals up to 20 times their size on benchmarks, offering strong performance without needing large hardware. The models are tailored for different hardware, from phones and IoT devices to personal computers and workstations. For example, the smaller E2B and E4B models are engineered for mobile and offline use with near-zero latency. This means developers can achieve capable AI with less hardware overhead.