Summary
Google DeepMind is rolling out Gemini 4 Argon, a new model that emphasizes coding and robotics, and this video offers a hands-on look at how it behaves. Access is available to Pro and Ultra subscribers in the Gemini app and AI Mode in search, while use in Google’s Anti-Gravity development platform requires the Ultra tier.
The presenter first tries the demos shown by Demis Hassabis, including a polished Wizard’s Chess game and a desk racing game built from a photo of a desk. He then tries building his own games, such as a micro RC car racer and a first-person sailing game, and finds the results less refined than those from other leading models. He attributes part of the gap to the model’s unusual steerability and says better prompting may close it.
The video also gathers reactions from Ethan Mollick and others who found the model’s style distinctly different from other frontier systems. A closing section covers Google’s claims for Argon as a physical reasoning model across the robotics lifecycle, including zero-shot robot control, real-to-sim reconstruction, and video world models, with context from Andon Labs’ Vending-Bench and Drone Bench evaluations.
📺 Source: Wes Roth · Published October 14, 2026
🏷️ Format: Review
