Google launched its latest and most advanced foundation model, Gemini 3, on Tuesday, making it immediately available through the Gemini app and AI search interface. This release, seven months after Gemini 2.5, positions Gemini 3 as Google's most capable large language model to date and an immediate contender among leading AI tools.
The introduction of Gemini 3 occurs amid accelerated development in frontier models, following OpenAI's GPT 5.1 and Anthropic's Sonnet 4.5. The model has already registered record scores on independent benchmarks, reinforcing its reported reasoning capabilities. Tulsee Doshi, Google's head of product for the Gemini model, stated, "With Gemini 3, we're seeing this massive jump in reasoning. It's responding with a level of depth and nuance that we haven't seen before."
On the Humanity's Last Exam benchmark, designed to assess general reasoning and expertise, Gemini 3 achieved a score of 37.4, marking the highest on record and exceeding the previous high of 31.64 held by GPT-5 Pro. The model also secured the top position on LMArena, a human-led benchmark that evaluates user satisfaction.
A research-intensive version of the model, named Gemini 3 Deepthink, is scheduled for release to Google AI Ultra subscribers in the coming weeks, pending further safety testing. Google reported that its Gemini app currently supports over 650 million monthly active users, with 13 million software developers leveraging the model as part of their operations.
Alongside the core model, Google also released Google Antigravity, a Gemini-powered coding interface. This new tool facilitates multi-pane agentic coding, combining a ChatGPT-style prompt window with a command-line interface and a browser window capable of showing the impact of code changes. DeepMind CTO Koray Kavukcuoglu commented that the agent is designed to "work with your editor, across your terminal, across your browser to make sure that it helps you build that application in the best way possible."
