Google has released Gemma 4 There are three sizes: 1B, 13b, and 27B, as well as a dense version of 31B. All Apache 2.0 protocols. Can be used for commercial purposes at will. This license change is more important than the model itself. Previously, Gemma used Google's own protocol with limitations. Now Apache 2.0 is directly competing with Meta's Llama. The highlight of the model itself: multimodal - text+visual+audio. The 31B dense version ran 89.2% on AIME 2026, while LiveCodeBench v6 achieved 80%, Codeforces ELO 2150。 The 27B parameter count is very friendly for local deployment. A 4090 can run. E4B and E2B are end side versions designed for mobile phones and IoT. Google is laying out the ecosystem for Gemini Nano 4. Llama has been dominant in the open source LLM community for too long. This time, Google is not testing the waters, but launching a comprehensive attack - covering the entire parameter range from 2B to 31B, with no commercial threshold for the Apache protocol, and integrating both the end and cloud. This is good news for independent developers and small teams. The more intense the competition, the more free lunches there will be. For Meta, Llama's moat is narrowing.