Google Unveils Gemini 3: A Leap Forward in AI Reasoning and Interactive User Interfaces
On November 18, 2025, Google introduced Gemini 3, its latest advancement in artificial intelligence, designed to transform how users interact with AI by bringing their ideas to life. The initial release, Gemini 3 Pro, is now available through the Gemini app and AI Mode.
Evolution of Gemini Models
The Gemini series has seen significant progress since its inception. Gemini 1.0 emphasized native multimodality and extended context windows. Subsequent versions, such as Gemini 2.0 and 2.5, introduced advanced reasoning, agentic capabilities, and enhanced coding functionalities. Gemini 3 continues this trajectory, focusing on understanding user intent with minimal prompting and delivering nuanced, insightful responses.
Enhanced Reasoning and Contextual Understanding
Gemini 3 excels in grasping the depth and nuance of user queries, enabling it to perceive subtle clues in creative ideas and dissect complex problems effectively. This advancement positions Gemini 3 as a true thought partner, capable of translating intricate scientific concepts into high-fidelity visualizations and facilitating creative brainstorming sessions.
Benchmark Achievements
Gemini 3 Pro has set new standards across various benchmarks:
– LMArena Score: Achieved a score of 1501, surpassing the previous leader, Gemini 2.5 Pro, which scored 1451.
– Humanity’s Last Exam: Demonstrated PhD-level reasoning with a top score of 37.5% without tool assistance.
– GPQA Diamond: Attained a score of 91.9%, showcasing exceptional general-purpose question-answering capabilities.
– MathArena Apex: Set a new standard in mathematics with a state-of-the-art score of 23.4%.
– Multimodal Reasoning: Achieved breakthrough scores of 81% on MMMU-Pro and 87.6% on Video-MMMU, highlighting its proficiency in processing diverse data types.
– Factual Accuracy: Scored 72.1% on SimpleQA Verified, indicating significant progress in delivering accurate information.
– Web Development: Topped the WebDev Arena leaderboard with an impressive ELO score of 1487.
– Tool Use Ability: Scored 54.2% on Terminal-Bench 2.0, reflecting its capability to operate computer terminals effectively.
– Coding Proficiency: Outperformed previous models on SWE-bench Verified with a score of 76.2%, measuring its effectiveness as a coding agent.
These achievements underscore Gemini 3 Pro’s ability to solve complex problems across various domains with high reliability.
Introduction of Deep Think Mode
Google also announced the upcoming Gemini 3 Deep Think mode, which offers enhanced reasoning and multimodal understanding. This mode outperforms Gemini 3 Pro on benchmarks like Humanity’s Last Exam (41.0% without tool use) and GPQA Diamond (93.8%). It also achieves an unprecedented 45.1% on ARC-AGI with code execution, demonstrating its ability to tackle novel challenges. Deep Think mode will be available in the coming weeks for AI Ultra subscribers.
Generative User Interfaces
A standout feature of Gemini 3 is its ability to create generative user interfaces (UIs). This capability allows the AI to generate entire user experiences, including web pages, games, tools, and applications, tailored to specific user prompts. For instance, when explaining the microbiome to a child, Gemini 3 can generate content and features appropriate for that age group, differing from those designed for adults. Similarly, it can create a gallery of social media posts for a business or generate a plan for an upcoming trip, each with a customized interface.
This functionality is launching as experiments in the Gemini app:
– Dynamic View: Designs and codes fully customized interactive responses for each prompt, providing a tailored experience.
– Visual Layout: Creates immersive, magazine-style views complete with photos and modules, enhancing the visual appeal and interactivity of the content.
Initially, users may experience one of these experiments to allow Google to gather feedback and refine the features.
Integration with Google Search and AI Mode
For the first time, a new model is being integrated into Google Search and AI Mode alongside the Gemini app. Starting this week, AI Pro and AI Ultra subscribers can select Thinking: 3 Pro reasoning and generative layouts from the dropdown menu in the top-left corner. This integration enhances the search experience by performing additional, nuanced searches to improve the final response.
AI Mode will also utilize generative UIs to create interactive tools and simulations. For example, users might access a mortgage calculator that allows adjustments to interest rates and down payments or engage with a physics simulation when learning about specific topics.
Gemini 3 will soon be available to all AI Mode users in the U.S., with subscribers receiving higher usage limits. In the coming weeks, Google plans to update Search’s automatic model selection to route challenging questions to Gemini 3 while using faster models for simpler tasks.
Introduction of Google Antigravity
Alongside Gemini 3, Google announced Google Antigravity, a new agentic development platform that enables developers to operate at a higher, task-oriented level. This integrated development environment (IDE) allows agents to work across the editor, terminal, and browser. Available now on Mac, Windows, and Linux, it leverages Gemini 3, Gemini 2.5 Computer Use, and Nano Banana.
With Google Antigravity, agents can autonomously plan and execute complex, end-to-end software tasks simultaneously on behalf of the user while validating their own code.
Conclusion
The launch of Gemini 3 marks a significant milestone in AI development, offering advanced reasoning capabilities and the ability to generate customized user interfaces. These innovations promise to enhance user interactions with AI, making them more intuitive and tailored to individual needs. As Gemini 3 continues to roll out, users can look forward to a more dynamic and responsive AI experience.
