Google releases Gemini 4 Argon for physical reasoning in robotics
11/20/2026 — 11/20, 11:16·1 sources (all)·1 reports (all)
Key takeaways
- Google released Gemini 4 Argon, a general-purpose physical reasoning model that supports robot control, Real2Sim scene reconstruction, video…
Background & analysis
According to an InfoQ Chinese report published at 19:00 on November 20, 2026, Google has released Gemini 4 Argon, a general-purpose physical reasoning model built for embodied AI tasks. It supports robot control, Real2Sim scene reconstruction, video annotation, and iterative policy optimization, rather than serving as a narrow end-effector policy. On the Embodied Reasoning QA (ERQA) benchmark, the model reached an 87% score using pure reasoning without any auxiliary tools, and Google stated that this result leads other frontier models on that test.
In the prompts that have been published, Argon was trained in a MuJoCo twin environment and then guided Boston Dynamics' Spot robot through a foot pedal task. It can manipulate relatively simple robotic arms using visual coordinates and HTTP motion primitives, and it enables closed-loop control through the Gemini Omni video world model. The Real2Sim examples reconstruct MuJoCo scenes from teleoperation videos, and annotation prompts label demonstration videos to support behavior cloning. Google also synthesized a domain-randomized MuJoCo scene from a single RGB photo and used it to fine-tune a vision-language-action policy, reporting that the success rate on the corresponding real-world task rose from 56% to 75%. Google did not disclose the failure rates of each demonstration.
The model is now available to users through the Gemini API and Google AI Ultra, with the model ID gemini-4-argon.
AI-generated from 1 reports · updated 23 hours ago
Related AI tools
- Heat index
- 32
- All sources
- 1
- All reports
- 1
- First seen
- 2 days ago
Reports on this story headlines open the original
InfoQ China reports that Google released Gemini 4 Argon, a general-purpose physical reasoning model covering robot control, Real2Sim scene reconstruction, video annotation and policy optimization, scoring 87% on the ERQA benchmark without tool use. Google says real-world task success rose from 56% to 75%, though it did not publish failure rates for its demos. The model is available through the Gemini API and Google AI Ultra under the ID gemini-4-argon.
InfoQ 中文AI score 69
Other stories people are talking about
- Heat index 133Google adds GIF generation to Gemini, powered by Omni 1.2 Flash2 sources
- Heat index 66Google opens Gemini Tasks agent mode to free users in six countries1 sources
- Heat index 64Google Gemini Black Friday: AI Pro annual plan about $99.991 sources
- Heat index 33Google Rebrands Spark as Tasks, Shifts to Gemini 4 Hydrogen1 sources
- Heat index 84Google Search and Gemini launch Connected Apps for shopping and tax tasks2 sources
How is heat calculated?About the methodHide
Heat counts how many independent sources covered a story in the last 48 hours: one source counts once no matter how many posts it published, decaying with a 24-hour half-life. What ranks first is what many people are talking about.
This page aggregates public feeds. Headlines and summaries are machine-organized and remain the property of the original authors; verify important facts at the source. If you believe a headline or summary infringes your rights, email the contact address in the footer with the page URL and basis of your claim, and we will remove or replace it after review.
- Surge
- Discussion rising fast
- New
- First report within 6 hours
- Rising
- Still gathering discussion

