On Tuesday, Google released Gemini 3, its newest and most advanced foundational model. This is now readily available through the Gemini app and AI search interface.
Arriving just seven months after the release of Gemini 2.5, this new model is Google’s most capable LLM to date and an immediate contender for the most capable AI tool on the market. This release also comes less than a week after OpenAI released GPT 5.1 and just two months after Anthropic released Sonnet 4.5, reminding us of the breakneck pace of frontier model development.
A more research-intensive version of this model, called Gemini 3 Deepthink, will also be available to Google AI Ultra subscribers in the coming weeks after passing further safety testing.
“Gemini 3 represents a huge leap forward in inference,” said Tulsee Doshi, head of product for Gemini models at Google. “They’re responding with a level of depth and nuance that I’ve never seen before.”
Some of its reasoning power has already been registered in independent benchmarks. The model scored 37.4, marking the highest score ever on the Humanity’s Last Exam benchmark, which aims to capture general reasoning and expertise. The previous highest score was 31.64 held by GPT-5 Pro. Gemini 3 also topped the leaderboard on LMArena, a human-driven benchmark that measures user satisfaction.
According to Google, the Gemini app currently has more than 650 million monthly active users, and 13 million software developers use the model as part of their workflow.
In addition to the base model, Google also released a Gemini-powered coding interface called Google Antigravity. This enables multi-pane agent coding similar to agent IDEs such as Warp and Cursor 2.0. Specifically, Antigravity combines a ChatGPT-style prompt window with a command line interface and a browser window that allows you to view the effects of changes made by your coding agent.
tech crunch event
san francisco
|
October 13-15, 2026
“The agent works with the editor, device, and browser to help you build your application in the best way possible,” said Koray Kavukcuoglu, CTO at DeepMind.
