Chinese Academy of Sciences' Maxwell model tops Meta-World benchmark with 91.9
Maxwell, an embodied AI model from the Chinese Academy of Sciences’ Institute of Artificial Intelligence for Industries, achieved a record 91.9 score on the Meta‑World benchmark, which tests robots on 50 everyday physical tasks such as grasping, opening doors, and using drawers. The model also completed more than 200 tasks without additional fine‑tuning and reached a 99.1% success rate in the…
Key points
- Maxwell scored 91.9 on Meta‑World, the highest ever recorded on the 50‑task robot benchmark.
- The model runs on 1 billion parameters with single‑card inference, handling over 200 tasks without fine‑tuning and 99.1% success in LIBERO.
- Competing submissions placed FabriVLA second (90) and SUREFlow third (88.3), while Chinese models PhysBrain and Unitree’s 6 B model also showed strong results.
Second‑place FabriVLA, developed by Shenzhen‑based Youibot, scored 90, while South Korea’s SUREFlow placed third with 88.3. Chinese models continue to close the gap with Western rivals: DeepCybo’s PhysBrain earned 72.5 on a suite of 28 benchmarks, just behind OpenAI’s GPT‑6 Astra at 73.3. Unitree Robotics released a 6‑billion‑parameter foundation model that can handle over 60 whole‑body and tabletop tasks, planning to reinvest nearly half of its IPO proceeds into robot‑intelligence research.
Model page: GPT-6 Astra →
A new Chinese embodied model tops global AI ranking on physical tasks scoring 91.9 on the Meta-World benchmark
scmp.com · 3 October 2026
Loading the full article…
This text was published by scmp.com and written by Zhang Tong. It is reproduced here with attribution so you can read it in full; the rights remain with the publisher. Read it at the source ↗
Coverage and discussion
1source- Reddit discussionreddit.com
The headline, key points and digest above were generated by Digest AI's editorial model from the linked sources. Automated summaries can contain errors: the sources are the record. Spotted a mistake? Tell us. Published by Martin K., who runs Digest AI and handles corrections.
More in Robotics & Physical AI
All →- Clone Robotics shows teleoperated hand, targets November launch for Torso 4 · 1 src
- Omron discusses next-generation LD mobile robots on Robot Report podcast · 1 src
- Eli Lilly and Purdue share HRI insights at RoboBusiness 2026 · 1 src
- F.02 Decommission · 1 src
- GMEX Robotics’ Hartzband outlines US expansion and social AI focus · 1 src
Comments
via GitHub Discussions