Gemini 3 is here: What you should know (www.techradar.com)

Google has rolled out Gemini 3, a major upgrade to its family of large models that’s being baked into the Gemini app, Search, and developer tools. It ships in two modes—Gemini 3 Pro (the full-featured, widely available version) and Deep Think (an enhanced-reasoning mode currently in testing for Google AI Ultra subscribers)—and Google is routing harder queries to Gemini 3 immediately for Pro/Ultra users. The release emphasizes improved reasoning over raw scale and aims to make AI a more practical part of everyday Google workflows rather than an experimental novelty. Technically, Gemini 3 intensifies multimodality and long-context capability: a million-token context window lets it hold sprawling sessions, combine inputs (e.g., handwritten recipes plus voice notes) into coherent outputs, and its video analysis now understands movement and timing (usable for things like sports analysis and training plans). The Gemini app gets a redesigned UI, “My Stuff” for generated content, real-time generative interfaces, and an agent that can execute multi-step tasks across connected Google apps. In Search, Gemini 3 builds generative, visual answers—tables, diagrams, interactive simulations and custom calculators—and uses a fan-out approach plus automatic model selection to find material earlier versions missed. The net effect: richer, app-like answers and deeper automation, with access initially gated to paid tiers as Google tests the most advanced capabilities.