Gemini 4 Argon: physical reasoning and simulation for robotics
Gemini 4 Argon is a physical reasoning model for robotics that scores 87% on the Embodied Reasoning (ERQA) benchmark. It creates MuJoCo digital twins, controls robots in closed visual loops, and synthesizes simulation data.
What changed
Gemini 4 Argon applies general-purpose physical reasoning across the robotics stack, scores 87% on ERQA, and supports MuJoCo twins, closed-loop control, teleoperation-to-sim transfer, and simulation data synthesis.
Why it matters
Developers and researchers can build simulation twins, generate domain-randomized data, and improve real robot success rates from 56% to 75% while reducing completion time from 35s to 24s.
Who should care
DevelopersResearchersEnterprises
Worth trying
Yes — Available on Google's developer platforms; supports MuJoCo, simulation and control.