Google Showcases Gemini 4 Argon Capabilities with Antigravity

Google demonstrated the capabilities of its new flagship model, Gemini 4 Argon, in conjunction with the Antigravity environment. The model is capable not only of writing code but also of independently bringing projects to a finished result.

Model Demonstration

• Animation of a pelican riding a bicycle using Fourier epicycles, where Argon handles physics and rendering, while Lyria 3.5 generates music.

• Stylized games with 3D rendering on Three.js, created from a single prompt.

• Multi-pass canvas renderer in Python for turning photographs into oil paintings.

• Interactive Mahjong on WebGL with realistic spring physics and sound effects.

• 3D musical instruments using WebMIDI and Three.js.

Technical Specifications and Availability

The Gemini 4 Argon model was announced on September 30, and its rollout began in the Antigravity desktop application and CLI for Google AI Ultra subscribers on October 13. For business use, the model is available via the Gemini API and as part of Gemini Enterprise subscriptions. Antigravity builds for Argon now offer context window options: 256K, 512K, and 900K tokens.

Separately worth noting is the Lyria 3.5 model, responsible for audio. It generates full-length tracks with verse, chorus, and bridge structures in 44.1 kHz stereo, accepting text and images as input.

Before the public launch, Google tested the models internally: it audited changes to access control systems, resolved tickets from routine rotations, and localized hundreds of documentation pages into 17 languages.

The company notes that when working with multiple tools, models are trained to respect system boundaries, honestly report negative results, and request human confirmation before performing irreversible actions.