Google backs open source models in Silicon Valley’s stance against Anthropic

Silicon Valley is splitting into two camps over the future of AI, and the fault line runs right through the concept of openness. On one side, Google, Meta, Microsoft, and Nvidia are doubling down on open-weights models that anyone can download, modify, and deploy. On the other side, Anthropic and OpenAI are lobbying for tighter controls, particularly on Chinese open-source AI, arguing that safety risks demand a more cautious approach. Google just made its position unmistakable. The Gemma 4 family from Google DeepMind, launched on April 2, 2026, represents the company’s most ambitious open contribution to the AI ecosystem yet. Released under the permissive Apache 2.0 license, the models span multiple sizes, including E2B, E4B, 26B MoE, and 31B dense variants, and Google claims they deliver the best intelligence-per-parameter of any open models available worldwide. The numbers tell the adoption story By early April 2026, Gemma had surpassed 400 million downloads and spawned more than 100,000 community-created variants. The Gemma series has been building toward this moment since its first model dropped in February 2024. Gemma 2 followed later that year, Gemma 3 arrived in 2025, and no...