Google has officially introduced Gemini 3, a significant enhancement to its leading multimodal AI model. This upgraded version promises improved reasoning capabilities and enhanced multimodal functionality, enabling it to work seamlessly across various formats such as voice, text, and images while operating in an agent-like manner.
The previous iteration, Gemini 2.5, allowed users to input multimodal data but often required specific instructions to generate the desired output, typically defaulting to plain text. With Gemini 3, Google has implemented what it terms “generative interfaces,” which empower the model to autonomously determine the most suitable output format. For instance, when users request travel recommendations, Gemini 3 can create a web-like interface within the app that includes interactive modules, images, and personalized follow-up questions to enhance user engagement. Similarly, when explaining concepts, the model can produce diagrams or animations to facilitate understanding, showcasing its ability to generate more visually engaging content.
In addition to these features, Google is rolling out Gemini Agent—an experimental tool designed for handling multi-step tasks directly within the app. This agent can integrate with services like Google Calendar and Gmail, enabling it to perform tasks such as organizing emails or managing schedules efficiently. The agent breaks down complex tasks into manageable steps, providing real-time progress updates and prompting user approval before proceeding. Google envisions this feature as a move towards creating a more comprehensive generalist agent. Starting November 18, Gemini Agent will be available on the web for Google AI Ultra subscribers in the U.S. Moreover, Gemini 3 will deepen integration with existing Google products, allowing a select group of AI Pro and Ultra subscribers to access a reasoning-focused version of the model for enhanced search capabilities, as well as generating personalized shopping recommendations based on Google’s vast Shopping Graph, which boasts over 50 billion product listings. Overall, these advancements signal a notable leap forward in AI interaction, promising a more intuitive and effective user experience.
Source: Google’s new Gemini 3 “vibe-codes” responses and comes with its own agent via MIT Technology Review