Google has unveiled Gemini 3, its most advanced AI model to date, calling it a “massive jump” in intelligence and reasoning. The new system replaces Gemini 2.5 and is now live across Google Search’s AI Mode and the Gemini app.

The company says the upgrade brings smarter responses, deeper context understanding, and entirely new ways of interacting with information.

Google CEO Sundar Pichai described Gemini 3 as a major advancement in how AI interprets intent, with the model now “much better at figuring out the context and intent” behind user requests. This means fewer prompts, more accuracy, and responses that prioritize insight over flattery.

The company says Gemini 3 is now the “best model in the world for multimodal understanding,” capable of processing and generating text, images, code, and interactive visuals with far greater depth than previous versions.

Gemini 3 Pro rolls out first

Google’s release begins with Gemini 3 Pro, now available “in preview.” The model significantly outperforms Gemini 2.5 Pro across all major AI benchmarks. According to Google, it delivers responses that are “smart, concise, and direct,” bringing “a new level of depth and nuance” to interactions.

Next in line is Gemini 3 Deep Think, currently undergoing safety testing. Deep Think, designed for more complex reasoning, will launch later for Google AI Ultra subscribers.

Improved reasoning, coding

Pichai noted that in just two years, AI has gone from “reading text and images to reading the room.” Gemini 3 reflects that evolution with stronger analytical skills and more human-like interpretation of user intent.

The model can now:

Convert information across formats
Generate visualizations and apps from a single prompt
Solve advanced science and math problems
Deliver interactive graphics in Google Search
Support deeper coding workflows

Google DeepMind CTO Koray Kavukcuoglu called Gemini 3 “our most intelligent model,” capable of helping people “bring any idea they have to life.”

Gemini 3 is now powering the most complex queries in Google Search and AI Mode, while smaller Gemini variants will handle simpler tasks. The model is also embedded across the Chrome browser, the Gemini app, and Google Cloud.

The rollout reflects Google’s strategy to showcase faster returns on its years-long AI investment by integrating new models across its ecosystem at once.

New developer platform

Alongside Gemini 3, Google introduced Antigravity, a development platform that enables AI-driven autonomous coding agents. These agents can write, test, and verify code across an editor, terminal, and browser—effectively functioning as automated software builders.

In a demonstration, Google showed an agent creating a full flight-tracking web application by coordinating tasks across tools.

Gemini 3 Deep think for advanced reasoning

Gemini 3 Deep Think tests multiple hypotheses simultaneously, picking the most accurate response after evaluating several reasoning paths. The system is designed for multistep tasks such as coding, research, and strategic planning.

It will be available exclusively to Google AI Ultra subscribers, the company’s premium tier priced at $249.99 per month.

Enhanced safety and security

Google says Gemini 3 is its most secure model so far. It has been trained to resist prompt injection attacks—attempts to trick the AI into ignoring safety controls or leaking sensitive information.

This improvement comes as the company increases AI integration across core products and user-facing tools.

AI driving growth across Google cloud

Gemini’s expansion is already contributing to Google’s financial performance. In the last quarter:

Google Cloud revenue grew 34% year-over-year
Revenue reached $15.2 billion, beating expectations
About 70% of Google Cloud customers now use Google’s AI tools

Executives called Gemini one of the company’s strongest engines of growth.

As part of its expansion strategy, Google will offer US college students free access to Gemini Pro for one year.