Overview
- Gemini 3 Pro is now live in preview in the Gemini app for all users, with AI Mode in Search available to Google AI Pro and Ultra subscribers in the US.
- Google cites benchmark leadership, including an Elo 1501 on LMArena, strong results on Humanity’s Last Exam and GPQA Diamond, and top multimodal scores on MMMU‑Pro and Video‑MMMU.
- New generative interfaces can produce magazine‑style layouts, dynamic views, simulations, and custom interactive tools directly in responses across the app and Search.
- An experimental Gemini Agent begins limited rollout for multi‑step tasks, while the new Antigravity platform lets developers run agentic coding across editor, terminal, and browser.
- A higher‑reasoning Deep Think mode remains limited to safety testers, with access planned for AI Ultra subscribers after evaluations.