The agent finally got a body. rpent, open-sourced september 21 by tsinghua's rlinf team, runs…
the agent finally got a body. rpent, open-sourced september 21 by tsinghua's rlinf team, runs robots on gpt-6 astra with 92.6% task success on libero-pro and 7x faster end-to-end task completion.
the frontier moved out of the chat window.
Context
The RLinf/RPent repository describes RPent (Recursive Physical Agent) as an open framework for building embodied agents, created 7 July 2026, listing Claude Code and Codex as agentic planners and Pi0.5 as a VLA, with LIBERO as the supported simulator. A first-party lab post of 21 September 2026 says it was jointly initiated by Tsinghua University, Infinigence and a third company, open-sourced that day and built by the core RLinf team. Secondary outlets report 92.6 percent on LIBERO-PRO (741 of 800 with a Codex planner on GPT-6 Astra, against 11.0 percent for pi0.5 alone) and about 7x faster task completion in Flash Mode on LIBERO Object (283.6 seconds to 40.9).
The identity and release date are first-party, and by Tsinghua's RLinf team is a partial description. The 92.6 percent and the 7x are secondary only, are lab-reported, and are two different measures: the 7x is an execution-time comparison in one mode on one suite and not end-to-end completion across LIBERO-PRO. Both are simulator results. Runs robots is not shown for real hardware, since no real-robot result was read, and the lab post says GPT-6 Astra is only beginning to accept real robot operation tests. The 11 percent baseline compares against the same VLA without the agent wrapper.
Watch next
- A first-party leaderboard or paper with the 741 of 800 figure and any real-robot demo.
Sources
- RLinf/RPent (GitHub)github.com
- RPent release post (Tsinghua lab, 21 Sep 2026)thusigs-edi-lab.github.io
- Tsinghua RPent embodied agent infrastructure (The Robo Radar, 21 Sep 2026)theroboradar.com
Provenance
The note above is reproduced unedited from the original post, first published on Threads on 21 September 2026 at 15:54 IST. Sources are the papers and datasets the note draws on.
View the original post ↗Embed this note
More notes
The air is now being asked to keep its own ledger
the air is now being asked to keep its own ledger: ecmwf’s aifs compo becomes the first ai model to forecast atmospheric composition globally every three hours, cleanair simulates 365 days of pm2.5 over china in ten seconds, and a unified framework maps six pollutants at one kilometer across the whole country. the air now files its own composition report.
read the note →The current is now being asked to draw its own map
the current is now being asked to draw its own map: china’s langya 2.0 predicts six ocean phenomena including internal waves and mesoscale eddies, a deep net called wenhai resolves eddies globally with air sea flux formulas built in, and scripps infers surface currents from the way temperature patterns deform in satellite images. the ocean now files its own circulation report.
read the note →The soil is now being asked to report its own carbon
the soil is now being asked to report its own carbon: a nix color sensor paired with generative data augmentation predicts soil organic carbon without a lab, random forest drives 74 percent of soil health mapping studies, and sentinel 2 tracks five year carbon change across france and italy from 922 samples. the dirt now files its own carbon account.
read the note →