Google launched Gemma 4 on April 2, 2026. The release covers four open-weight AI models. For the first time in the Gemma series, every model ships under the Apache 2.0 open-source license.
The licensing change is the most consequential part of the announcement. Previous Gemma releases carried restrictions that blocked or complicated commercial use.
Apache 2.0, a permissive open-source standard widely used across the software industry, grants developers near-total freedom to modify, distribute, and deploy the models in commercial products without royalty obligations. That single policy shift transforms Gemma 4 from a research artifact into a production-ready tool for American startups and enterprise developers alike.
The four models span a range of hardware targets, from smartphones and Internet of Things (IoT) devices at the low end to server-grade deployments at the high end. The smallest models are built for offline use, capable of running multimodal AI tasks, processing both text and images simultaneously, without a network connection on consumer devices.
Context windows, the amount of text a model can process in a single pass, reach 128,000 tokens on edge-optimized variants and scale up to 256,000 tokens on the larger models, according to ph.investing.com.
Gemma 4's Benchmark Standing and Gemini Roots
The models share a research and technology foundation with Gemini 3, Google's proprietary flagship AI system,. That lineage gives Gemma 4 capabilities that earlier open-weight releases from any company rarely matched.
The 31-billion-parameter (31B) variant ranks among the top open models globally on the Arena AI leaderboard, a widely cited third-party evaluation platform. Performance on mathematical reasoning and coding benchmarks is also reported as strong, per kursol.io. The full model family adds advanced reasoning and multimodal processing to the open-source AI ecosystem.
The release arrives at a moment of diverging philosophies across the AI industry. Alibaba Group Holding, the Chinese technology conglomerate, has moved in the opposite direction, shifting from open-source AI releases toward paid enterprise models.
In the United Kingdom, regulators launched an investigation into Microsoft's AI-cloud licensing practices over concerns about market dominance. Google's decision to open Gemma 4 under Apache 2.0 fits a different strategic posture, one that positions the company as a permissive alternative to closed or increasingly commercialized ecosystems. Gemma 4 is available through Google's AI developer platforms as of the release date.
