Google Gemini Now Generates Interactive 3D Models and Simulations for Users

Text answers are no longer the ceiling. Google has begun rolling out a feature for its Gemini chatbot that converts plain-text questions into fully interactive 3D models and simulations, rendered directly inside the chat window, where users can rotate geometry, adjust sliders, and change variables in real time.

The capability runs on the Pro-tier Gemini model, accessed through the Gemini web app. Users select the Pro model in the prompt bar and phrase requests as "show me" or "help me visualize" to trigger the outputs. Under the hood, the system generates live WebGL and Three.js code, the same rendering stack Google's Android XR team used for an immersive blood-cell biology simulation.

The practical ambition is substantial. An early benchmark documented by R&D World Online found Gemini producing a working 3D model of GLP-1 receptor activation, complete with 4,000 animated lipid particles and cinematic lighting, in approximately two minutes from a single text prompt. The output was fully rendered HTML and Three.js code. Google also showcased Gemini coding a 3D journey through the scale of the universe, from a proton to the observable universe, a demonstration it described as a significant leap in on-demand generative coding.

That same R&D World Online test introduced a pointed caveat: the GLP-1 model "prioritizes aesthetics over molecular perfection," with the publication flagging the artificial flatness of the rendered lipids. That gap between visual persuasiveness and scientific fidelity matters most in classroom and research contexts. Separate benchmarking by Artificial Analysis found that Gemini 3 Pro, despite achieving the highest overall accuracy of any model tested at 53 percent, registered an 88 percent hallucination rate, matching its predecessors. Google has not specified what accuracy flags or disclosure mechanisms accompany simulation outputs, placing the burden of domain verification on the user.

The feature is the public-facing product of a roadmap stretching back to December 2024, when Google introduced Genie 2, capable of generating playable 3D worlds from a single image. Genie 3, released subsequently, extended that to fully interactive, physics-aware environments built in real time from text or image prompts. Google DeepMind frames the long-term goal as building a "world model" that can "make plans and imagine new experiences by understanding and simulating aspects of the world, just as the brain does."

Education is the most structured deployment target. At ISTE 2025, Google announced interactive diagrams were coming to Gemini for educational users, beginning with students 18 and older before expanding to younger age groups. Educators are already constructing "Interactive Simulations" Gems, custom AI experts grounded on specific course assignments, within the Gemini app.

On the competitive front, 3D design platform Spline has integrated Gemini 2.5 Pro Preview into its Hana canvas editor, allowing users to generate full interactive 2D and 3D experiences from text prompts. Gemini 3's underlying performance is supported by a 1,487 Elo score atop the WebDev Arena leaderboard and a 76.2 percent result on SWE-bench Verified, a benchmark measuring AI coding agent quality that directly underpins how well generated 3D interfaces actually function.

Google AI Ultra subscribers gain a premium variant through Deep Research, where reports now include interactive simulation models that let users adjust variables and forecast outcomes within the document itself. The base interactive 3D feature is rolling out to all Gemini app users globally, making the accuracy question not just an enterprise concern but a mainstream one.