At Google I/O 2025, the company officially announced Gemini 3 Pro — a new flagship language model that, according to internal tests, significantly outperforms Gemini 2.5 Pro across all major benchmarks. Alongside it, Google introduced Antigravity — a free platform for creating and running AI agents.
What Gemini 3 Pro Can Do
Google emphasized that its priority was not polished phrasing but real usefulness: the model now understands user intent more accurately, interprets contextual nuances more deeply, and provides information that aligns more directly with what the user needs. In Search’s AI Mode (available through the dropdown menu), responses have become noticeably more precise and interactive.
The preview version is already available:
A separate mode, Gemini 3 Deep Think, demonstrated outstanding performance in the most challenging tests: the “Final Exam of Humanity,” GPQA Diamond (natural sciences), and ARC-AGI-2. This mode is currently restricted to safety testers, but Google stated that it will be available to Google AI Ultra subscribers in the coming weeks.
Antigravity — An IDE for AI Agents
The second major announcement was Antigravity, a fully free (public preview) development environment where users can create AI agents capable of performing tasks autonomously, writing code, and verifying it. Google described it as essentially a version of VS Code “supercharged” with built-in Gemini agents that operate on behalf of the user.
In Short
Google released Gemini 3 Pro — its most powerful and advanced model to date. It is already available in the U.S. and to Google AI subscribers, with broader limits expected soon. The company also launched the free Antigravity platform for building AI agents. The year 2025 is shaping up to be Google’s strongest moment yet in the large-language-model race.
month
week
day