Executive Summary
Google has announced Gemma 4, its most advanced family of open-source models, built with technology from the Gemini 3 program. The release includes four distinct sizes optimized for hardware ranging from mobile devices to developer workstations, focusing on advanced reasoning, code generation, and agentic workflows. By releasing Gemma 4 under a permissive Apache 2.0 license, Google aims to provide the developer community with state-of-the-art, efficient AI tools that are both powerful and widely accessible.
Key Takeaways
* Four Versatile Sizes: The family includes an Effective 2B (E2B) and 4B (E4B) for on-device tasks, a 26B Mixture-of-Experts (MoE) model for low-latency inference, and a 31B Dense model for maximum quality.
* State-of-the-Art Performance: The 31B model is ranked as the #3 open model globally on the Arena AI leaderboard, demonstrating high "intelligence-per-parameter" and outperforming models 20x its size.
* Advanced Capabilities: All models are designed for agentic workflows with native support for function-calling and structured JSON output. They also feature advanced reasoning, code generation, and long context windows (up to 256K).
* Multimodal & Multilingual: The models natively process video and images. The E2B and E4B versions also support audio input for speech recognition. The entire family is trained on over 140 languages.
* Permissive Apache 2.0 License: In response to community feedback, Gemma 4 is released under a commercially permissive open-source license, giving developers complete control over their data, infrastructure, and models.
* Immediate Availability: Model weights are available immediately on platforms like Hugging Face, Kaggle, and Ollama, with extensive support across the AI ecosystem, including Google AI Studio, Vertex AI, NVIDIA NIM, and more.
Strategic Importance
This launch solidifies Google's commitment to the open-source AI community, directly competing with other leading open models. By offering highly efficient, state-of-the-art models under a permissive license, Google aims to drive broad adoption and foster an ecosystem that complements its proprietary Gemini models and encourages use of its cloud infrastructure.