In the early hours of this morning, Google CEO Sundar Pichai took to social media to announce a game-changing development: the launch of Gemini 3, alongside an entirely new development platform called Google Antigravity — Google's answer to the increasingly popular Cursor IDE.
"Gemini 3 is the world's best multimodal understanding model," Pichai declared in his post. "It's also our most powerful agent and code generation model yet. Gemini 3 quickly understands what you're thinking and what you need — just describe it simply, and you'll get exactly what you want."
In an interesting show of industry camaraderie, OpenAI CEO Sam Altman chimed in with congratulations: "Congrats to Google on launching Gemini 3! Looks like a great model."
Two Flavors of Gemini 3
Google is rolling out two versions of its latest AI powerhouse: Gemini 3 Pro and Gemini 3 Deep Think.
Gemini 3 Pro: The Standard Bearer
Gemini 3 Pro represents Google's next-generation flagship model, bringing substantial improvements across multiple dimensions.
Enhanced Reasoning Capabilities
The model demonstrates impressive reasoning prowess. Take this example: upload a photo of a handwritten recipe in any language, and Gemini 3 can identify it, translate it into your preferred language, and generate a complete digital recipe — all in one seamless operation. This involves image recognition, text extraction, language translation, and content reorganization, and Gemini 3 handles it effortlessly.
Comprehensive Multimodal Understanding
Gemini 3 can simultaneously process text, images, video, and code. Consider this real-world application: you're playing pickleball and want to improve your game. Upload your match footage to Gemini 3, and it will analyze your technique, identify areas for improvement, and create a personalized training plan.
This capability extends to educational contexts as well. Upload academic papers or video lectures, and Gemini 3 will generate interactive study cards or visualizations to help you better grasp the material.
Outstanding Code Generation
Gemini 3's performance in code generation is particularly impressive. On the WebDev Arena leaderboard, it achieved an Elo rating of 1487, placing it at the top of the rankings.
In the SWE-bench Verified benchmark test — specifically designed to evaluate code agent capabilities — Gemini 3 scored 76.2%. On Terminal-Bench 2.0, which tests a model's ability to operate computers through terminal commands, it achieved 54.2%.
Gemini 3 Deep Think: For Complex Challenges
Deep Think mode is designed for extraordinarily complex problems. According to Google, this version is undergoing additional safety evaluations and will be available to Google AI Ultra subscribers in the coming weeks.
Google Antigravity: A New Era of AI-Powered Development
Beyond Gemini 3 itself, Google introduced a significant new product: Google Antigravity, a next-generation development platform specifically designed for agent-based development.
Traditional IDEs serve as tools for developers. Antigravity aims to go further — to become a "proactive partner" in the development process. Within Antigravity, AI agents are elevated to a first-class position, with direct access to the editor, terminal, and browser.
These agents can autonomously plan tasks, execute complex end-to-end software development work, and validate their own code.
Core Capabilities of Antigravity
Antigravity integrates multiple models working in concert. Gemini 3 Pro serves as the core, handling advanced reasoning and code generation. Gemini 2.5 Computer Use handles browser control. Additionally, there's Nano Banana (Gemini 2.5 Image), an image editing model.
This combination enables Antigravity to handle more comprehensive development tasks than traditional tools.
Real-World Application Scenarios
At the launch event, Google demonstrated a compelling use case. A developer wanted to create a flight tracking application. They simply described their requirements to Antigravity.
Antigravity's agent sprang into action: first planning the entire project, then writing the code, testing the application in the browser, and finally verifying that all functionality worked correctly. The entire process was automated, with the developer only needing to provide confirmation and guidance at key decision points.
This workflow allows developers to focus on higher-level task design while AI handles the implementation details.
Getting Started with Gemini 3
Gemini 3 began rolling out on November 18th.
For General Users
You can access Gemini 3 through the Gemini app. Google AI Pro or Ultra subscribers can also use it in Search's AI Mode.
For Developers
Developers can access Gemini 3 through multiple platforms:
- Google AI Studio
- Google Antigravity
- Gemini CLI
- Vertex AI
The Bigger Picture
Google shared some impressive statistics at the launch:
- AI Overviews reaches 2 billion users monthly
- The Gemini app exceeds 650 million monthly users
- Over 70% of Google Cloud customers are using AI features
- 13 million developers are using Google's generative models
- Gemini 3 features a context window of 1 million tokens
The launch of Gemini 3 marks another significant milestone in Google's AI journey. From a technical standpoint, it brings substantial improvements across multiple dimensions. From a product perspective, Google Antigravity opens up new possibilities for AI-assisted development.
For everyday users, Gemini 3 means smarter search and more capable applications. For developers, Gemini 3 and Antigravity mean more efficient development workflows and more powerful tools. The AI IDE space just got a formidable new competitor — and the race to redefine software development is heating up.