Google has officially unveiled Gemini 3, marking what the company calls its “most intelligent model” and a major leap forward in its journey toward advanced artificial intelligence. The launch signals a new chapter for Google’s rapidly expanding Gemini ecosystem, arriving simultaneously across Search, the Gemini app, AI Studio, Vertex AI, and the company’s newly announced agentic development platform, Google Antigravity.
In a statement, Google and Alphabet CEO Sundar Pichai described Gemini 3 as a milestone that brings the company’s full AI stack together from infrastructure to research to global-scale products. “Every generation of Gemini has been built on the last,” he said, adding that Gemini 3 represents Google’s ability to deliver advanced capabilities faster than ever. The model brings major improvements in reasoning, context understanding, and multimodal processing, enabling it to “read the room” rather than simply read text or images.
Stronger Reasoning and Multimodal Intelligence
Developed by Google DeepMind, Gemini 3 arrives with immediate claims of state-of-the-art performance.
The flagship Gemini 3 Pro dramatically outperforms its predecessor across all major AI benchmarks, topping the LMArena leaderboard and earning top marks on high-level evaluations such as Humanity’s Last Exam and GPQA Diamond. Its multimodal reasoning capabilities also set new standards in image, video, and spatial understanding.
A new enhanced reasoning mode, Gemini 3 Deep Think, pushes performance even higher, offering unprecedented results on some of the toughest AI challenges. Deep Think will undergo additional safety reviews before rolling out to Google AI Ultra subscribers.
Built to Learn, Build and Plan
Google is positioning Gemini 3 as a universal AI assistant designed to help users “learn anything, build anything, and plan anything.”
Learning: Gemini 3 can combine text, images, video, and code to create personalised learning experiences, from translating handwritten family recipes into digital cookbooks, analysing sports footage and generating coaching plans.
Building: For developers, Gemini 3 brings richer web UI generation, advanced agentic coding, and top performance on WebDev Arena and SWE-bench. The model is now integrated into major developer tools and platforms, including AI Studio, Vertex AI, Gemini CLI, and third-party services like GitHub and Replit.
Planning: Gemini 3 also showcases major improvements in long-horizon reasoning, enabling it to autonomously carry out extended, multi-step workflows such as inbox organisation or service booking under user supervision.
A highlight of the launch is Google Antigravity, a new agent-first development environment. By giving Gemini agents direct access to the editor, browser, and terminal, Antigravity enables AI to independently plan, code, and validate complex software tasks.
A Focus on Responsible AI
Google says Gemini 3 is its most secure model to date, undergoing extensive testing for cybersecurity, prompt injection resistance, and sycophancy reduction. The company collaborated with global experts, including UK AISI and independent evaluators like Vaultis and Dreadnode, to conduct safety audits ahead of release.
Rolling Out Now
Gemini 3 is available starting today in the Gemini app, AI Mode in Search (for Pro and Ultra subscribers), AI Studio, Vertex AI, and Gemini Enterprise. Gemini 3 Deep Think will follow after further safety evaluations. Google also confirmed that additional models in the Gemini 3 series will be announced soon.
With Gemini 3, Google aims to expand the limits of AI intelligence and usability. As Pichai concludes, “We hope you like Gemini 3, we’ll keep improving it, and we look forward to seeing what you build.”
