Third place on the benchmark. That was the first headline about Gemma 4. But a far more important fact was quietly buried beneath it: for the first time, Google attached an Apache 2.0 license to an AI model. It could not unseat the top two in the numbers race, but in the license war it changed the board itself.
What Actually Happened: Four Sizes, One Decisive Change
Google unveiled Gemma 4 on April 2, 2026. It comes in four versions by parameter count: 2B, 4B, 26B, 31B. The largest 31B model placed 3rd on the Open LLM Global Arena and achieved 89.2% on the AIME 2026 math benchmark. Multimodal capabilities were also substantially upgraded. It natively processes video, image, and audio input and was trained on more than 140 languages. The design covers the full range from edge devices to enterprise servers. Yet the most decisive change of all these figures came down to just two words: Apache 2.0.
Why This Matters More Than People Think: The License Is the Strategy
Previous Gemma models were distributed under Google's own custom license. That license placed several restrictions on commercial use and required review by corporate legal teams. Apache 2.0 is different. Modification, redistribution, and commercial use are all free, with no separate negotiation required. By comparison, Meta's Llama models use a custom license that includes a clause requiring a separate commercial license once monthly active users exceed 700 million. It seems like no immediate problem for a startup, but the moment that company grows or gets acquired by a large enterprise, legal risk arises. Gemma 4's Apache 2.0 has no such clause at all. It is a signal that Google intends to directly block Llama's solo run in the open-source AI ecosystem.
Hidden Insight: Google Is Copying Its Android Strategy Into AI
In 2007, Google released Android as open source. The goal was not to build the best smartphone OS. It was to put Google services on as many devices as possible. Gemma 4's Apache 2.0 strategy follows the same logic. More than the model itself, the more that Gemma-based apps and services spread across the world, the stronger Google's cloud (Google Cloud), developer tools (Vertex AI, Google Colab), and, over the long term, its advertising ecosystem become. While OpenAI and Anthropic stack up API revenue with closed models, Google is laying down an ecosystem with open source. Historically, this strategy has succeeded once before. Today, 72% of the world's smartphones run Android. Can Gemma become the next Android? That is the most important question in open-source AI in 2026.