Google DeepMind launches Gemini 4 Argon for Robotics, a physical reasoning model for robot development

Google DeepMind/Gemini Robotics

Google DeepMind released Gemini 4 Argon as a general-purpose physical reasoning model for the robotics lifecycle, from understanding demonstration video to generating robot code and simulation data. The post reports an 87% score on the Embodied Reasoning (ERQA) benchmark, a figure the vendor says beats every other frontier model. Almost all results are vendor-reported demos.

• Released 13 Oct 2026 as model ID gemini-4-argon via the Gemini API (paid tier) and Google AI Studio, with Google Antigravity support.
• Access to AI Studio and Antigravity is limited to Google AI Ultra subscribers.
• On the ERQA benchmark it scores 87% on pure reasoning, which the post says beats every other frontier model.
• In one simulation case it generated 4,500+ simulated episodes; real-world success rose from 56% to 75% and completion time fell from 35 s to 24 s.
• Caveats: no independent evaluation, no failure rates, the sim-to-real gains come from one task on one setup, and no pricing or regional availability is given.