Google Releases Gemma 4, Its Most Powerful Open AI Models Yet

Google has released Gemma 4, a new family of open artificial intelligence models that the company calls its most intelligent yet. The four models are built for advanced reasoning, agentic workflows, and on-device deployment, and they arrive under a fully permissive Apache 2.0 license that lets anyone use them for free in commercial projects.

What Makes Gemma 4 Different

The Gemma 4 lineup includes two larger models with 26 billion and 31 billion parameters designed for server-side tasks, plus two compact variants called E2B and E4B that are built to run on phones and laptops. All four use a Mixture-of-Experts architecture, a technique where only a small fraction of each model’s total parameters activate during any given task. This means a model with billions of parameters can deliver strong results while using far less memory and battery power. Google claims the compact models outperform competing artificial intelligence models up to 20 times their size on key reasoning and math benchmarks. The larger models support context windows of 256,000 tokens, roughly equivalent to a 500-page book, while the compact versions handle 128,000 tokens.

Built for Agents and Reasoning

Every Gemma 4 model ships with a built-in reasoning mode that lets it think step by step before answering, similar to how large language models like GPT-4 or Claude approach complex problems. The models also include native function calling, which is essential for building AI agents that can browse the web, query databases, or interact with external tools without extra engineering. Google says the models can process images and video at variable resolutions, and the smaller variants even support speech recognition for voice-driven applications. The shift to an Apache 2.0 license marks a major change from earlier Gemma versions that imposed more restrictive terms on commercial use. This move positions Google to compete directly with Meta’s Llama and other open-weight model families for developer loyalty across a rapidly growing AI ecosystem.

The release signals that the open-source AI race is accelerating fast. By giving away its best reasoning technology for free, Google is betting that developers who build on Gemma will eventually migrate to its cloud platform and broader AI stack, creating long-term revenue even without upfront licensing fees.