The American technology company Google has announced Gemma 4, a new generation of open models designed for advanced reasoning and agentic workflows. In their announcement, they share that it is built on the same technology as Gemini 3, the family includes four distinct sizes: Effective 2B (E2B) and Effective 4B (E4B) for mobile and IoT devices, along with a 26B Mixture of Experts (MoE) and a 31B Dense model for workstations and high-end GPUs. These models represent a significant leap in intelligence-per-parameter, with the 31B version already ranking as the #3 open model globally on the Arena AI leaderboard.
Key technical advancements in Gemma 4 include native multimodal support for vision, audio, and video, as well as expanded context windows of up to 256K tokens. The models are engineered for complex logic, multi-step planning, and autonomous agent tasks through native function-calling and structured JSON output. To support global accessibility, they are trained on over 140 languages and are released under a commercially permissive Apache 2.0 license.
The release emphasizes wide ecosystem compatibility, offering day-one support for platforms like Hugging Face, NVIDIA NIM, and Ollama. Android developers can also begin prototyping with the edge models via the AICore Developer Preview.
