Particle.news
Download on the App Store
51 ARTICLES
·
1h ago

Google Launches Gemini 3, Bringing Its New Flagship AI to Search, the Gemini App and Developers

The release channels benchmark-topping reasoning and coding into real products through generative interfaces, agent features and a new Antigravity development platform, with a deeper‑thinking variant pending safety review.

The Google Gemini 3 promotional image.
Gemini Agent in action in a Google demo video.
Image
Image

Overview

  • Gemini 3 Pro is live in the Gemini app for all users, while U.S. subscribers to Google AI Pro or Ultra can select “Thinking: 3 Pro” in Search’s AI Mode, with broader access planned.
  • Google cites leading results on LMArena (1501), WebDev Arena, GPQA Diamond (91.9%) and Humanity’s Last Exam (37.5%) alongside strong MMMU and MathArena scores and a 1 million‑token context window.
  • New generative interfaces let responses become interactive layouts, simulations, graphs or tools instead of plain text, rolling out as experiments in the Gemini app and in Search AI Mode.
  • An experimental Gemini Agent can connect to Gmail, Calendar and Reminders to perform multi‑step tasks with user approvals, and the new Antigravity IDE enables agentic, end‑to‑end coding workflows.
  • A higher‑reasoning Deep Think mode that outperforms Pro on key tests will reach AI Ultra subscribers after additional safety evaluations, while APIs are available today via AI Studio, Vertex AI and the Gemini API.