Google AI Studio has used Gemini 4 Argon to build an interactive 3D island sandbox. It runs in the browser, and its terrain, water, and sand all respond to physics in real time.
What Google actually showed
Gemini 4 Argon is Google’s newest AI model. Its pitch is generating intricate, interactive 3D experiences directly from user prompts.
Island Builder is the headline act. It works as a browser-based sandbox where physics simulations run live as users manipulate the environment.
It isn’t the only showcase. Google’s demonstrations also include Wizard Chess 3D, Desk Racer, and a range of shared 3D environments.
The technical foundation matters here. Gemini 4 Argon leans on improved spatial reasoning and visual coding. The model works with web graphics frameworks like Three.js and WebGPU, the tools that let browsers render 3D scenes without plugins.
The specs and the price tag
Gemini 4 Argon ships with a 1 million unified token context window.
On benchmarks, the model scored 77.9% on DeepSWE v1.1, a test focused on software engineering tasks.
Google has also laid out its pricing. Access through the Gemini API comes at an introductory rate of $2 per million input tokens and $10 per million output tokens. That pricing runs until November 30, 2026.
After that, rates move to the standard $4 per million input tokens and $20 per million output tokens.
How Argon got here
The rollout came in stages. Google first gave limited access to cybersecurity partners on September 30, 2026, through what it calls the Fairwind Program. Public launch followed on October 13, 2026.
Beyond 3D demos, Google positions Gemini 4 Argon for advanced software engineering and enterprise knowledge work. Its listed strengths include long-horizon autonomy and multimodality.
Disclosure: This article was edited by Diego Almada Lopez. For more information on how we create and review content, see our Editorial Policy.