Google Gemini Learns to Generate Interactive 3D Models and Simulations

Google has updated Gemini: the chatbot can now generate interactive 3D models and physics simulations directly in the conversation. Simply ask a question — and instead of static text or an image, the system builds an object that you can interact with in real time. This is one of the most notable expansions of response formats in large language models in recent months.

The function works as follows: in response to a request, Gemini creates a three-dimensional model with a set of interactive control elements. The user can rotate the object with the mouse, move sliders, enter numerical values — and immediately see how the behavior of the system changes. No downloads, transitions to third-party services, or waiting for rendering: everything happens inside the chat interface, without any additional tools.

This is a key difference from situations where a chatbot simply generated a link to an external simulation or offered to download a file. An editor from The Verge tested the function with a specific example. He asked Gemini to build a simulation of the Moon orbiting the Earth — without any code or prior setup.

The chatbot generated a three-dimensional scene with several control tools: a slider to adjust the orbital speed, a toggle to control the visibility of the orbital trajectory, and a pause button. The physics worked correctly, objects moved smoothly, and all control elements responded instantly. Such a result previously required the work of a programmer and specialized software.

Until now, generative chatbots primarily dealt with text, code, and images. Interactive simulations are a qualitatively different level: they require not just drawing an object, but generating working logic for its behavior, including physics and reaction to user input. In essence, Gemini now acts not only as a content generator, but as a constructor of educational and research tools on the fly, without involving developers.

The application of such a response format is obvious in several fields. In education: instead of reading a description of planetary orbits, a student can move a slider and see how changes in speed affect the trajectory. In engineering — quickly visualize a mechanism or physical process without running specialized software like MATLAB or SolidWorks.

In medicine — examine an anatomical structure from all angles. In design — evaluate the shape of an object in three dimensions right during the discussion, without leaving the conversation. Google has not yet revealed technical details about exactly how Gemini generates the logic of simulations — whether it writes code under the hood using WebGL, Three.

js, or uses a different approach. Neither the methodology nor the limitations on types of supported objects have been officially announced. However, from the perspective of the end user, implementation details are not essential: what matters is the result, and it already looks convincing at the stage of the first demonstrations.

This update fits into a broader competitive trend. Large AI laboratories are competing not only in the accuracy of text responses, but also in the richness of output formats. OpenAI launched Canvas for collaborative document and code editing, Anthropic develops artifacts with interactive HTML, and now Google is adding 3D and physics simulations.

Chatbots are gradually transforming from text assistants into multi-format work environments. For Gemini users, this means expanded capabilities without additional technical skills. Want to understand how a pendulum works — ask.

Want to see a molecule in three dimensions — ask that too. Want to compare the behavior of two different springs — set the parameters and watch. The boundary between a question and an interactive textbook continues to blur, and 3D simulations are one of the most vivid confirmations of this.