Google Says New Gemini 3 AI Model Will Better Understand Your Requests

Gemini 3, Google’s latest AI model, is the company’s most intelligent model yet, with multimodal capabilities and ambiance coding capabilities, the company said in a blog post Tuesday. It’s available now.

Google says Gemini 3 is “designed to capture depth and nuance” and better understand the intent behind a user’s request. The company also touted Gemini 3’s multimodal capabilities, such as its ability to turn a long video lecture into interactive flash cards or analyze a person’s pickleball game and find areas for improvement.

Gemini 3 is about more than just the app. It will also be available in AI mode in search and, for Pro and Ultra subscribers, in AI previews. In AI previews, Gemini 3 can generate interactive elements.

Don’t miss any of our unbiased technical content and lab reviews. Add CNET as your preferred Google source.

Demis Hassabis and Koray Kavukcuoglu of Google DeepMind said in a blog post that Gemini 3 Pro is less sycophantic, a problem that plagues AIs and leads to AI psychosis in some. It is also safer against rapid injection attacksin which bad actors attempt to cause an AI to ignore its original instructions and perform unintended actions.

The company also unveiled Google Antigravity, a new agent development platform. Google says Antigravity is like having an active partner when building tools or working on projects, autonomously planning and executing complex software tasks while validating your own code. It works in tandem with Gemini 2.5’s computer usage model for browser control and works with nano bananathe Gemini 2.5 image model.

Gemini 3’s new, more powerful agentic capabilities will only be available for $250/month Google AI Ultra subscribers initially. This will allow the Gemini agent to perform multi-step workflows, such as scheduling a travel itinerary.

Learn more: AI Essentials: 29 Ways to Make AI Generation Work for You, According to Our Experts

Google’s release of Gemini 3 comes as an AI war intensifies between it, OpenAI, Anthropic and xAI. Google has always topped the AI ​​charts, although other AI models aren’t far behind, sometimes trading places at the top. With the release of Gemini 3, Google seems to be trying to solve some of AI’s more annoying problems, like hallucinations or sycophancy. It is also about proving that AIs can be truly agents, capable of carrying out tasks on behalf of the user. Other agent models have proven problematic in real-world use and run into difficulties. various security issuesespecially in web browsers.

Google’s latest AI version also comes at a time when there are fears of an AI bubble forming in the stock market. AI companies including Nvidia, Google, Meta and Microsoft make up 30% of the S&P 500. Google is currently valued at $3.4 trillion. Even Google CEO Sundar Pichai says the trillion-dollar AI investment boom has “elements of irrationality” and that an explosion would affect all AI companies, he said in a BBC interview. Still, if Google wants to keep its stock price rising, it needs to demonstrate that its AI models beat the competition.