On Tuesday, Google launched Gemini 3, its latest and most superior foundational mannequin. That is now available by way of the Gemini app and AI search interface.
Arriving simply seven months after the discharge of Gemini 2.5, this new mannequin is Google’s most succesful LLM so far and a direct contender for probably the most succesful AI software available on the market. This launch additionally comes lower than per week after OpenAI launched GPT 5.1 and simply two months after Anthropic launched Sonnet 4.5, reminding us of the breakneck tempo of frontier mannequin improvement.
A extra research-intensive model of this mannequin, referred to as Gemini 3 Deepthink, may even be out there to Google AI Extremely subscribers within the coming weeks after passing additional security testing.
“Gemini 3 represents an enormous leap ahead in inference,” stated Tulsee Doshi, head of product for Gemini fashions at Google. “They’re responding with a stage of depth and nuance that I’ve by no means seen earlier than.”
A few of its reasoning energy has already been registered in unbiased benchmarks. The mannequin scored 37.4, marking the best rating ever on the Humanity’s Final Examination benchmark, which goals to seize common reasoning and experience. The earlier highest rating was 31.64 held by GPT-5 Professional. Gemini 3 additionally topped the leaderboard on LMArena, a human-driven benchmark that measures consumer satisfaction.
In response to Google, the Gemini app at the moment has greater than 650 million month-to-month lively customers, and 13 million software program builders use the mannequin as a part of their workflow.
Along with the bottom mannequin, Google additionally launched a Gemini-powered coding interface referred to as Google Antigravity. This allows multi-pane agent coding much like agent IDEs equivalent to Warp and Cursor 2.0. Particularly, Antigravity combines a ChatGPT-style immediate window with a command line interface and a browser window that permits you to view the consequences of adjustments made by your coding agent.
tech crunch occasion
san francisco
|
October 13-15, 2026
“The agent works with the editor, system, and browser that can assist you construct your utility in the easiest way potential,” stated Koray Kavukcuoglu, CTO at DeepMind.
