Google launches Gemini 3 with advanced reasoning

Google has introduced Gemini 3, a new artificial intelligence model that, according to the company, better understands the context, intent and nuances of a request. Its goal is to help you get useful answers with fewer instructions and tackle more complex tasks.

The launch arrives on November 18, 2025, and marks the first time Google has integrated a new version of Gemini into its search engine from day one. It will also be available in the Gemini app and for developers.

What Gemini 3 brings

Google describes Gemini 3 as its most intelligent model yet. The main improvement is not just answering questions, but better understanding what you are trying to achieve.

For example, a brief request can contain a tone, an intention or several implicit conditions. Gemini 3 aims to pick up on those signals and provide a more relevant response without requiring you to explain every detail step by step.

The model combines capabilities Google has developed in previous generations:

• Understanding text and images, known as multimodality.
• The ability to work with large amounts of information.
• Reasoning to analyze problems with multiple steps.
• Agent-oriented features, meaning systems that can carry out tasks based on goals.

The company says Gemini 3 can better recognize the nuances of a creative idea and separate the different parts of a difficult problem. This could translate into uses such as turning a description into a prototype, analyzing related documents or planning a task that requires several decisions.

Coming to Search and the Gemini app

The first major change for the general public is in Search. Gemini 3 is being added to AI Mode, the search mode designed to answer complex queries through reasoning and interactive experiences generated during the search.

In practice, this is aimed at questions that cannot be answered well with a list of links. For example, comparing several options using specific criteria or breaking a broad query into different parts to build a more complete answer.

Gemini 3 is also coming to the Gemini app. According to Google, the model better understands the context of a conversation and needs fewer instructions to follow the purpose of a request. That does not remove the need to review its answers, but it may reduce the work of explaining a task from scratch.

Tools for developers

The model is available from launch in AI Studio and Vertex AI, Google's platforms for building and deploying artificial intelligence applications. It is also integrated into Google Antigravity, a new development platform focused on agents.

For you, this means that features based on Gemini 3 may start appearing in Google services and third-party applications. For developers, it means direct access to a model with reasoning and multimodal understanding capabilities for building their own products.

Gemini's scale is already considerable

Google says that AI Overviews reaches 2 billion users per month and that the Gemini app exceeds 650 million monthly users. It also says that more than 70% of its Cloud customers use one of its artificial intelligence tools and that 13 million developers have built products with its generative models.

These figures were reported by the company and are not an independent measurement. Even so, they show where much of the next stage will play out: not just in the models, but in their integration with search engines, apps and tools that many people already use.

Gemini 3 represents the next step in Google's strategy: making AI reason better, understand more context and act within everyday products. What matters now is seeing how it performs outside demonstrations and which specific features reach each user, country and service.