Google has launched its latest AI model, Gemini 4 Argon, which enables the creation of interactive 3D environments directly from user prompts. The model utilizes spatial reasoning and visual coding to generate browser-based experiences, such as the Island Builder sandbox, which features real-time physics for terrain, water, and sand.
Beyond the Island Builder, Google showcased other applications including Wizard Chess 3D and Desk Racer. The model integrates with web graphics frameworks like Three.js and WebGPU to render 3D scenes without requiring plugins. In performance testing, Gemini 4 Argon achieved a score of 77.9% on the DeepSWE v1.1 benchmark for software engineering.
Access to the model is provided through the Gemini API. An introductory pricing tier of $2 per million input tokens and $10 per million output tokens is available until November 30, 2026. Following this date, the standard rates will increase to $4 per million input tokens and $20 per million output tokens.
The release followed a phased approach, beginning with limited access for cybersecurity partners via the Fairwind Program on September 30, 2026. A public launch occurred on October 13, 2026. Google intends to utilize the model for advanced software engineering and enterprise tasks, highlighting its multimodality and long-horizon autonomy.
