Download Link (Hugging Face)
https://huggingface.co/collections/google/gemma-4
This is a quote from the Google Korea blog.
Introducing Gemma 4, the most powerful open model in terms of performance per byte
https://blog.google/intl/ko-kr/company-news/technology/gemma-4-kr/
Today, Google is introducing Gemma 4, the most intelligent open model ever. Specifically designed for advanced inference and agent-based workflows, Gemma 4 delivers unprecedented levels of intelligence per parameter. This groundbreaking achievement was made possible by the vibrant participation of the community. Since Google released the first version, developers have downloaded Gemma over 400 million times and built a 'Gemmaverse' ecosystem consisting of over 100,000 variant models. Google has continuously sought directions to expand the possibilities of AI, reflecting the needs of innovators, and Gemma 4 is Google's answer. Gemma 4 is available under the Apache 2.0 license, allowing anyone to use it widely. ……
Google offers Gemma 4 in four sizes: Effective 2B (E2B), Effective 4B (E4B), 26B Mixture of Experts (MoE), and 31B Dense. The full model lineup supports more than just simple chat, including complex logical processing and agent-based workflows. Google's large models deliver state-of-the-art performance at each size, with the 31B model ranking third among open models on the Arena AI text leaderboard, the current industry standard, and the 26B model ranking sixth. Moreover, Gemma 4 outperforms models that are 20 times larger. This improved performance per parameter allows developers to achieve high levels of AI functionality with relatively fewer hardware resources.
[The rest of the content is omitted]
