Google launches Gemini 3 with new coding app and record benchmark scores

On Tuesday, Google launched Gemini 3, its newest and most superior basis mannequin, which is now instantly out there via the Gemini app and AI search interface.

Coming simply seven months after the Gemini 2.5 launch, the brand new mannequin is Google’s most succesful LLM but, and a right away contender for probably the most succesful AI device in the marketplace. The discharge additionally comes lower than every week after OpenAI launched GPT 5.1, and a mere two months after Anthropic launched Sonnet 4.5 — a reminder of the blistering tempo of frontier mannequin growth.

A extra research-intensive model of the mannequin, known as Gemini 3 Deepthink, may also be made out there to Google AI Extremely subscribers within the coming weeks, as soon as it passes additional rounds of security testing.

“With Gemini 3, we’re seeing this huge leap in reasoning,” mentioned Tulsee Doshi, Google’s head of product for the Gemini mannequin. “It’s responding with a degree of depth and nuance that we haven’t seen earlier than.”

A few of that reasoning energy is already registering on impartial benchmarks. With a rating of 37.4, the mannequin marked the very best rating on file on the Humanity’s Final Examination benchmark, meant to seize common reasoning and experience. The earlier excessive rating, held by GPT-5 Professional, was 31.64. Gemini 3 additionally topped the leaderboard on LMArena, a human-led benchmark that measures consumer satisfaction.

In accordance with Google, the Gemini app at the moment has greater than 650 million month-to-month energetic customers, and 13 million software program builders have used the mannequin as a part of their workflow.

Alongside the bottom mannequin, Google additionally launched a Gemini-powered coding interface known as Google Antigravity, permitting for multi-pane agentic coding much like agentic IDEs like Warp or Cursor 2.0. Particularly, Antigravity combines a ChatGPT-style immediate window with a command-line interface and a browser window that may present the affect of the modifications made by the coding agent.

“The agent can work together with your editor, throughout your terminal, throughout your browser to make it possible for it helps you construct that software in one of the best ways potential,” mentioned DeepMind CTO Koray Kavukcuoglu.