Google has launched Gemini 3, its most advanced AI model to date, promising major leaps in reasoning, multimodality and agentic capabilities while rolling it out across Search, the Gemini app and new developer tools.
Google has unveiled Gemini 3, calling it its “most intelligent model yet” and marking what CEO Sundar Pichai describes as “a new era of AI” for the company.
In a major upgrade to its flagship AI platform, Google is rolling out Gemini 3 across Search, the Gemini app, developer tools, and enterprise services – positioning the model as a leap forward in reasoning, multimodal understanding, and agentic capabilities.
Pichai said the release builds on two years of rapid AI progress, with Gemini now used monthly by 2 billion people through AI Overviews, 650 million users in the Gemini app, and 13 million developers.
Gemini 3 arrives with two tiers: Gemini 3 Pro, available today in preview, and Gemini 3 Deep Think, a more advanced reasoning mode coming soon following additional safety checks.
Google says Gemini 3 Pro significantly outperforms Gemini 2.5 Pro across benchmarks, topping the LMArena leaderboard and setting new records in science, maths and multimodal tasks.
Deep Think pushes those scores even higher, demonstrating AGI-style long-horizon problem solving with unprecedented accuracy.
A core focus is multimodality, with Gemini 3 having the ability to process text, images, video, audio and code with deeper context understanding.
Google has said it can translate entire handwritten family cookbooks, break down complex academic papers into interactive visualisations, or analyse sports videos to generate personalised training routines.
The launch also introduces Google Antigravity, an “agent-first” development environment where Gemini can autonomously write, test and validate code across editor, terminal and browser surfaces – a shift Google claims will redefine software creation.
Gemini 3 also now powers AI Mode in Search, bringing interactive reasoning, dynamic layouts and generative visualisations to Google’s core product on day one – the first time Google has shipped a new Gemini model directly into Search at launch.
Google emphasised that Gemini 3 is its “most secure model to date”, with enhanced protections against prompt injection, misinformation, and model manipulation.
Safety partners, including the UK’s AISI, participated in external evaluations.
With more Gemini 3 models coming soon – and Deep Think rolling out to Ultra subscribers in the weeks ahead – Google says it is entering “the next era” of AI: one defined by agents, reasoning and personalised intelligence at global scale.
Google unveils ‘a new era of AI’ with Gemini 3
