Google has introduced Gemini 3, marking a major step toward artificial general intelligence (AGI). The model is designed for multimodal understanding and advanced reasoning, outperforming previous versions across multiple AI benchmarks.
Gemini 3 Pro demonstrates PhD-level reasoning with top scores on Humanity’s Last Exam and GPQA Diamond, and sets new standards in mathematics, multimodal reasoning, and factual accuracy.
The model also delivers enhanced interactivity, richer visualizations, and concise, insightful responses. Gemini 3 Pro can interpret text, images, video, audio, and code, supporting tasks ranging from translating handwritten recipes into shareable cookbooks to analyzing sports performance and creating visual learning tools.
Learning, Building, and Planning with AI
An upgraded mode, Gemini 3 Deep Think, pushes reasoning and multimodal capabilities further. In tests, it outperformed Gemini 3 Pro on advanced benchmarks like Humanity’s Last Exam and ARC-AGI-2, demonstrating improved problem-solving for novel and complex challenges.
Gemini 3 is integrated across Google products to enhance learning, development, and planning. AI Mode in Search uses Gemini 3 for immersive visual layouts and interactive tools. Developers can leverage the model via Google AI Studio, Vertex AI, Gemini CLI, and third-party platforms like GitHub and Replit. Gemini 3 leads in coding benchmarks, scoring highly in tool use and agentic coding tests.
Google also launched Antigravity, an agentic development platform. Using Gemini 3, Antigravity allows developers to plan, execute, and validate complex software tasks autonomously. The model supports longer-term planning, consistently executing multi-step workflows for daily tasks and business simulations.
Connecting the Dots: Gemini Across Google
Gemini 3 has undergone extensive safety evaluations, showing improved resistance to misuse and prompt injections. Google collaborated with external experts and organizations to ensure robust safety standards. Deep Think mode will be released gradually to select users for additional safety review.
This launch builds on Google’s broader shift to Gemini: earlier, Google began rolling out Gemini for Home, replacing Google Assistant on Nest speakers and displays. On the TV front, Gemini also supplanted the Assistant on Google TV Streamer devices, enabling more natural, conversational interactions on the big screen.
And in mapping, Google is transforming Maps with Gemini-powered conversational AI to offer hands-free directions, live recommendations, and more intuitive navigation.
Gemini 3 is now available for Google AI Pro and Ultra subscribers, developers, and enterprises. Additional models and capabilities in the Gemini 3 series are planned, expanding access to advanced AI tools for learning, building, and planning across industries.