Google's Gemini 4 Argon scores 87% on embodied reasoning benchmark, tops all frontier models
jackyliang42 · x · 2026-10-14
Google released Gemini 4 Argon for Robotics, touting a step change in physical understanding, code generation, and visual-semantic reasoning toward "physical AGI."
Key points:
- On the Embodied Reasoning (ERQA) benchmark for 3D physical/spatial problem solving, Argon scores 87% using pure reasoning, outperforming every other frontier model.
- Planning in world models: write a plan, simulate action consequences with an omni world model, reflect on rollouts, and iterate into a video plan.
- Four application areas demonstrated: robot control (zero-shot mobile manipulation on Boston Dynamics Spot, e.g. building a MuJoCo digital twin of a foot-pedal trash bin to practice before real-world execution; closed-loop steering of video world models), Real2Sim (reconstructing calibrated simulation twins from robot camera feeds, egocentric video, or hand-drawn sketches), physical understanding (dexterity labeling, dense value functions, progress estimation), and research/self-improvement (action improvement via simulation and on-hardware iteration).
Related event: Google Launches Frontier Model Gemini 4 Argon Worldwide(277 posts)→
More from Embodied
- When will a humanoid robot team beat the World Cup champs? Reliability still at 80% on easy tasks — mathemagic1an · 2026-10-15
- ETH's Uncertainty-Aware World Model Makes Offline RL Work on Real Robots — breadli428 · 2026-10-15
- UPenn lab shows GPT-6 Astra can pilot FPV drones from raw images, no mapping stack needed — antoniloq · 2026-10-15
- Robotics Researcher: The Last Two Weeks Were an RT-X-Level Moment for Sim RL — KyleMorgenstein · 2026-10-15
- Why is my Oura ring talking to me like a therapist? — var_epsilon · 2026-10-15
- SONIC now runs on full-size humanoids: Agibot A3 and UBTech Walker C1 — zhengyiluo · 2026-10-15