Google releases Gemma 4, a family of open models built off of Gemini 3 (www.engadget.com)

Google has launched Gemma 4, a new suite of open-weight models inspired by the advancements of its proprietary Gemini 3 Pro. This release introduces four versions tailored for different use cases: the 2-billion and 4-billion "Effective" models for edge devices, and the more powerful 26-billion "Mixture of Experts" and 31-billion "Dense" systems for robust machines. Notably, Google claims these models deliver an impressive "intelligence-per-parameter" ratio, with the two larger variants achieving top rankings on Arena AI's text leaderboard, surpassing models with significantly greater parameter counts. The Gemma 4 models are versatile, capable of processing images, video, and audio, making them suitable for various applications, including optical character recognition and speech understanding. Additionally, they can generate offline code, enhancing their usability in environments without internet access. By releasing Gemma 4 under an Apache 2.0 license, Google grants developers increased flexibility to modify and deploy these models across different environments, promoting digital sovereignty. This strategic move is expected to accelerate innovation and adoption within the AI/ML community, as developers gain greater control over their projects using these advanced tools, available on platforms like Hugging Face, Kaggle, and Ollama.