Overview
- Gemini 3 is live across core products, including Google Search and the Gemini app, with executives noting this is the first time a new model shipped into Search on day one.
- Google claims a record benchmark score of 1,501 and PhD‑level reasoning on exams such as Humanity’s Last Exam and GPQA Diamond, with multimodal handling of text, images, audio and code.
- New offerings include Gemini Agent for multi‑step tasks and Antigravity in preview for autonomous coding agents that operate across an editor, terminal and browser.
- Deep Think, a mode for parallel hypothesis testing on complex problems, launches first for Google AI Ultra subscribers at $249.99 per month, while US university students get Gemini Pro free for one year.
- Google says the model is its most secure yet with training against prompt‑injection attacks, and reports AI is driving revenue as Google Cloud sales rose 34% year over year to $15.2 billion.