Google officially launches Gemini 3: Vibe Coding, Deep Think, and a new era for agents

Gemini 3 is here. In a massive blog post published just today, Google CEO Sundar Pichai and DeepMind CEO Demis Hassabis introduced the new model family, calling it their “most intelligent model” to date. But this isn’t just a standard speed bump; Google is claiming that in just two years, their AI has evolved “from simply reading text and images to reading the room.”

Unlike previous launches where we had to wait weeks for the good stuff, Google is shipping Gemini 3 at scale starting today. Here is an overview of everything announced, from the new “Pro” and “Deep Think” models to the sci-fi-sounding “Antigravity” platform for developers.

Gemini 3 Pro: The new daily driver

The star of the show is Gemini 3 Pro, which replaces the current 2.5 Pro model. It is rolling out today in the Gemini App and in “AI Mode” within Google Search.

Google isn’t being shy about the performance gains here. They claim Gemini 3 Pro significantly outperforms Gemini 2.5 Pro on every major AI benchmark.

• #1 on the Leaderboards: It has already hit the #1 spot on the LMArena Leaderboard with a breakthrough score of 1501 Elo.
• Smarter, not nicer: One of the most interesting updates is a behavioral tweak. Google says Gemini 3 Pro trades “cliché and flattery for genuine insight — telling you what you need to hear, not just what you want to hear.”
• “Vibe Coding”: Google is leaning hard into “Vibe Coding.” The model tops the WebDev Arena leaderboard and is capable of generating richer, interactive web UIs (visualizations, simulations) from complex prompts. It scores 76.2% on SWE-bench Verified, a massive leap over the previous model.

Gemini 3 “Deep Think”: The reasoning monster

While “Pro” is for fast, daily tasks, Google is also introducing Gemini 3 Deep Think. This is their answer to the recent wave of “reasoning” models (like OpenAI’s o1) that take time to “think” before responding.

This model is designed for PhD-level complexity. In testing, it crushed the “Humanity’s Last Exam” benchmark with a 41.0% score and achieved an unprecedented 45.1% on ARC-AGI-2 (a benchmark for solving novel challenges). “Deep Think” is currently in the hands of safety testers. It will be available to Google AI Ultra subscribers in the coming weeks.

Real-world “Multimodal” Magic

We hear “multimodal” a lot, but Google gave some wild examples of what Gemini 3 can actually do by combining vision, text, and reasoning:

• Family Recipes: It can decipher handwritten recipes in different languages and translate them into a digital cookbook.
• Sports Coaching: You can upload a video of your pickleball match, and it will analyze your form and generate a training plan to fix your swing.
• Visual Learning: In Google Search, it can now generate interactive UIs on the fly—like a simulation of “how RNA polymerase works”—instead of just showing you a diagram.

For Developers: Introducing “Google Antigravity”

Perhaps the most surprising part of the announcement is a brand new product called Google Antigravity. Google calls this an “agentic development platform.” It sounds like a futuristic IDE (Integrated Development Environment) where the AI isn’t just a chatbot in the sidebar—it is the developer. In Antigravity, AI agents have dedicated access to:

1. The Code Editor
2. The Terminal
3. The Browser

This allows the agent to autonomously plan, code, and validate its own work. You can ask it to build an app, and it will write the code, run the terminal commands, and check the output in the browser to make sure it works. It even integrates the new “Nano Banana” (Gemini 2.5 Image) model for generating assets on the fly.

Where you can use it now

Google is moving fast with this rollout. As of today (Nov 18):

• Regular Users: Gemini 3 Pro is rolling out in the Gemini App (personal accounts) and AI Mode in Search (for AI Premium subscribers).
• Developers: It is available now in Google AI Studio and Vertex AI.
• Enterprise: Available in Gemini Enterprise.
• Deep Think: Coming soon to AI Ultra.

This is a massive competitive leap for Google. If you have a workspace or personal account, go check the Gemini app—you might already be running on Gemini 3.