Google has officially launched Gemini 3, which the company touts as its most intelligent model yet. CEO Sundar Pichai declared Gemini 3 to be "the best model in the world for multimodal understanding". Rolling out starting today, November 18, 2025, Gemini 3 combines advanced reasoning, revolutionary multimodal capabilities, and powerful agentic features to help users bring any idea to life.

Here is an in-depth look at what makes Gemini 3 Pro a monumental step forward for consumers and developers, incorporating key features that enhance its position as a frontier AI model.

State-of-the-Art Reasoning and Unprecedented AI Benchmarks

Gemini 3 is built on a foundation of state-of-the-art reasoning, enabling it to grasp depth and nuance, whether perceiving subtle creative clues or dissecting complex, multi-layered problems. The model is designed to better interpret the context and intent behind a request, meaning users can get what they need with less prompting.

Gemini 3 Pro significantly outperforms previous models on every major AI benchmark.

Key performance highlights include:

• LMArena Leaderboard: Gemini 3 Pro tops the leaderboard with a breakthrough score of 1501 Elo, surpassing its predecessor, Gemini 2.5 Pro.

• PhD-Level Reasoning: It achieved top scores on challenging academic tests, including 37.5% on Humanity’s Last Exam (without tools) and 91.9% on GPQA Diamond.

• Multimodal Reasoning: It achieved breakthrough scores of 81% on MMMU-Pro and 87.6% on Video-MMMU, redefining multimodal reasoning.

• Coding Capabilities: Gemini 3 Pro tops the WebDev Arena leaderboard (1487 Elo) and greatly outperforms 2.5 Pro on the coding agent benchmark, SWE-bench Verified (76.2%).

Pushing the Frontiers: Gemini 3 Deep Think

For tackling the most complex problems, Google introduced Gemini 3 Deep Think mode, an enhanced reasoning mode designed to push performance even further. In testing, Deep Think demonstrated superior performance, scoring 41.0% on Humanity’s Last Exam and 93.8% on GPQA Diamond. This enhanced reasoning mode is currently undergoing additional safety evaluations and will be available to Google AI Ultra subscribers soon.

The Context Revolution: 1 Million Tokens

One of the foundational strengths of Gemini 3 is its immense context window, supporting 1 million tokens. This capability significantly expands the scope of tasks the model can handle within a single request.

In practice, this capacity allows the model to process massive inputs like:

• Approximately 50,000 lines of code.

• The equivalent of eight average-length English novels.

• Large corpuses of text for summarization or complex question answering.

This capability is essential for agentic workflows, as it allows the model to reliably keep state and maintain information about the world and its goals.

Transforming User Experience with Generative UI

Gemini 3 makes possible Generative UI (User Interfaces), representing a major shift in how users interact with Search and the Gemini app. This capability allows the model to generate entirely customized, interactive user experiences—not just text output—including web pages, tools, simulations, and applications, generated automatically in response to a prompt.

This is rolling out in AI Mode in Search and the Gemini app as experimental features:

• Dynamic View: Customizes the user experience by designing and coding a unique, interactive layout tailored perfectly for the specific prompt, understanding that explaining a complex topic to a child requires a different layout and content than explaining it to an adult.

• Interactive Tools: In Search's AI Mode, Gemini 3 can generate interactive tools and simulations on the fly, such as a physics simulation or a mortgage calculator, allowing users to interact directly within the search results.

• Visual Layout: Creates an immersive, magazine-style view complete with photos and modules that allow users to customize results further using sliders or checkboxes.

The Developer Leap: Introducing Google Antigravity

Gemini 3 is positioned as Google's most powerful agentic and vibe coding model yet. To harness this power, Google released Google Antigravity, a new agentic development platform and AI-first Integrated Development Environment (IDE).

Google Antigravity is designed to allow developers to operate at a higher, task-oriented level. Key features that redefine the developer workflow include:

• Autonomous Agents: Agents within the platform can autonomously plan and execute complex, end-to-end software tasks across the editor, terminal, and browser, while simultaneously validating their own code.

• Multimodal Integration: Antigravity leverages Gemini 3’s multimodal capabilities to process code, natural language, images (like pasted screenshots of desired UIs), and API responses in real time.

• Zero-Config API Testing: The platform executes real-time, zero-config API testing directly inside the IDE, automatically generating requests and validating responses against schemas.

• Superior Context: The 1 million token context window means Antigravity can understand entire monorepos without truncation, allowing for accurate refactoring of massive codebases on the first attempt.

Accessibility and Frontier Safety

Google is committed to a responsible launch, confirming that Gemini 3 is its most secure model to date, having undergone the most comprehensive set of safety evaluations. Safety assessments confirmed that Gemini 3 Pro did not reach any Critical Capability Levels (CCLs) across the risk domains defined in the Frontier Safety Framework (FSF), including CBRN (chemical, biological, radiological, and nuclear information risks) and cybersecurity.

Access to Gemini 3 Pro

Gemini 3 Pro is rolling out immediately across the Google ecosystem:

• For Consumers: Available in the Gemini app and in AI Mode in Search for Google AI Pro and Ultra subscribers. Users can select 'Thinking' in the model drop-down to access Gemini 3 Pro.

• For Developers/Enterprises: Access is available through the Gemini API in AI Studio, Vertex AI, and the new Google Antigravity development platform. It is also available in third-party platforms such as Cursor, GitHub, and JetBrains.

If previous AI models were like having a powerful calculator or a highly efficient intern for specific tasks, Gemini 3 is like acquiring a full creative and technical team that can understand ambiguous instructions and autonomously build the solution, from drafting the blueprint to executing the final code.