Google Gemma 4

Gemma 4 provides a family of open‑source language models designed for advanced reasoning and agentic workflows. The models are available in four sizes—Effective 2B, Effective 4B, a 26B mixture‑of‑experts, and a 31B dense variant—allowing developers to run sophisticated AI capabilities on personal computers, mobile devices, or IoT hardware. They are built on the same research foundation as Gemini 3 and are distributed under an Apache 2.0 license, emphasizing accessibility and openness.

The larger models achieve high rankings on the Arena AI text leaderboard, with the 31B model placed third among open models and the 26B model sixth, demonstrating strong performance relative to their parameter counts. The smaller Effective models prioritize low‑latency, multimodal processing for on‑device use, offering a “mobile‑first” AI experience while maintaining the ability to handle complex logic beyond simple chat.

Gemma 4 targets developers who need powerful language capabilities that can be deployed locally, reducing reliance on large‑scale cloud resources. Its open‑source licensing and compatibility with a variety of hardware make it suitable for experimentation, research, and building applications that require high‑quality reasoning without proprietary constraints.