Google Releases Gemini 3 AI Models

Google announced the release of Gemini 3 on Tuesday, its latest suite of artificial intelligence models designed to advance reasoning, coding, and multimodal capabilities. The initial offering, Gemini 3 Pro, is now available in preview across consumer products like the Gemini app and Google Search's AI Mode, as well as developer tools including Google AI Studio and Vertex AI. This marks the first time Google has integrated a new AI model into Search on launch day, aiming to deliver more nuanced and interactive responses to users.

Gemini 3 Pro outperforms its predecessor, Gemini 2.5 Pro, on major benchmarks, achieving top scores such as 1,501 on the LMArena leaderboard for user satisfaction and 1,487 ELO on the WebDev Arena for coding tasks. It also scored 37.5 on Humanitys Last Exam without tool use, demonstrating PhD-level reasoning in complex subjects like science and mathematics. Google describes the model as its most intelligent yet, with a one-million-token context window that allows it to handle extensive information while providing concise, direct responses that prioritize accuracy over flattery. An enhanced version, Gemini 3 Deep Think, which improves scores to 41 on...