Why Google held back a huge new AI model at its big conference

At this year's I/O conference, Google CEO Sundar Pichai surprised attendees by announcing that the much-anticipated Gemini 3.5 Pro AI model would not be unveiled as expected. The news was met with disappointment, highlighting the competitive landscape in AI development. Pichai hinted that the delay is strategic, allowing Google to refine the model's capabilities in coding tasks. The AI landscape is rapidly evolving, with competitors like Anthropic's Claude Code and OpenAI's Codex making significant strides. These players have captured developer interest and revenue by providing innovative AI tools that streamline coding processes. Instead of the flagship model, Pichai focused on the Gemini 3.5 Flash, a more compact and cost-effective alternative. Despite being slightly less powerful, this model is already integrated into Google's Antigravity AI coding service, which launches today for software developers. This tool promises to assist in code generation while collecting valuable feedback data. For example, if developers encounter issues that halt their projects, Google can analyze this feedback to enhance the Gemini 3.5 Pro model. The reinforcement learning approach will enable the AI to learn from successful and unsuccessful coding outputs, providing critical insights for future improvements. Pichai expressed excitement about the forthcoming 3.5 Pro model, assuring the audience that it would significantly enhance coding capabilities when it is ready next month. The anticipation builds as developers await this powerful addition to the AI coding toolkit.