In a decisive move to reclaim the AI throne from OpenAI’s GPT-5 Pro, Google yesterday officially unveiled Gemini 3, a new generation of models that the company claims will shift the industry from “chatbots” to “autonomous agents.”
Launched globally on November 18, Gemini 3 drops the decimal point from previous versions (like 1.5 or 2.0), signaling a complete architectural overhaul aimed at “vibe coding”, complex reasoning, and a new paradigm Google calls Generative UI.
“Gemini 3 is not just a model; it’s the engine for the agentic era,” said Google CEO Sundar Pichai. Unlike its predecessors, Gemini 3 is designed to execute multi-step workflows—such as researching a topic, writing code, and deploying a web app—without constant human hand-holding.
The launch introduces a new tiered lineup:
- Gemini 3 Pro: Available immediately to all users via the Gemini App and the new “AI Mode” in Google Search. It is described as the “workhorse” for daily tasks.
- Gemini 3 Deep Think: A new “reasoning-heavy” model designed to compete with OpenAI’s o1/o2 series. It will be available to Gemini Advanced (Ultra) subscribers in the coming weeks.
- Gemini Agent: An experimental tool for Ultra subscribers that can browse the web and control third-party tools to complete end-to-end errands.
The most buzzworthy announcement is Google Antigravity, a new developer platform (IDE) designed specifically for “agentic software development.”
Vibe Coding: Google describes this as a coding style where developers describe the intent and feel of an app, and Gemini 3 writes, debugs, and deploys the code autonomously.
Generative UI: In a massive shift for Google Search, Gemini 3 can now generate interactive visual interfaces on the fly. If a user asks for a mortgage calculator or a travel itinerary, Gemini 3 doesn’t just return text—it builds a custom, clickable widget or mini-app instantly.
According to Google DeepMind, Gemini 3 Pro has achieved a score of 37.5% on “Humanity’s Last Exam,” a notoriously difficult benchmark for general reasoning, edging out OpenAI’s GPT-5 Pro (which scored 31.6%). It also reportedly tops the leaderboard on ARC-AGI with a 45.1% score when allowed to use code execution.
“We are moving from models that read text to models that ‘read the room’ and act on it,” said Demis Hassabis, CEO of Google DeepMind.
Gemini 3 is live now for users worldwide. Developers can access the API via Google AI Studio starting today. The specialized Deep Think model and the Antigravity IDE are currently in preview for select partners and will roll out broadly by December.
For more Tech and AI news, stay tuned to MyMobileTrend
