Summary (tl;dr)

Google DeepMind has released Gemma 4, its newest family of open-source artificial intelligence models, designed for advanced reasoning and multimodal tasks, making them freely available under an Apache 2.0 license for widespread use.

Essential Background

The Gemma family of models originated in February 2024 as lightweight, open models built from the same research and technology as Google's Gemini models, with subsequent versions Gemma 2 and Gemma 3 released in June 2024 and March 2025, respectively. While earlier Gemma versions permitted local use and modifications, their licensing was more restrictive, not fully open-source. The initial Gemma models were a strategic move by Google to contribute to the open-source AI community, allowing developers to build and customize AI applications.

The Full Story

Today, Google DeepMind unveiled Gemma 4, the latest generation of its open-weight AI models, now fully open-source under the permissive Apache 2.0 license. This release significantly enhances capabilities for advanced reasoning, agentic workflows, and multimodal understanding, processing text, image, video, and audio inputs. Gemma 4 comes in four distinct sizes (E2B, E4B, 26B A4B, and 31B), enabling deployment across a wide range of hardware, from high-end phones and laptops to servers and powerful workstations. The larger models, particularly the 31B variant, have already achieved high rankings on industry-standard AI leaderboards, outperforming models significantly larger in size.

Why It Matters

The release of Gemma 4 under a truly open-source license is a significant development for the AI community, fostering greater innovation and accessibility. It allows developers and enterprises complete control over their data, infrastructure, and models, enabling secure deployment across various environments. The optimized performance of Gemma 4 models on diverse hardware, including mobile devices, democratizes access to state-of-the-art AI capabilities, potentially leading to a new wave of localized AI applications and personal AI assistants. The inclusion of advanced reasoning and agentic workflow features also empowers developers to create more sophisticated and autonomous AI systems.

Geographic Location

• Virtual/Online (Google DeepMind's announcement of Gemma 4)
• Mountain View, Santa Clara County, California, United States (Google's headquarters, where DeepMind's research and development efforts are coordinated)