A preview version of AgiBot's next robot manipulation foundation model has taken the top spot on RoboDojo, a public benchmark for generalist robot manipulation policies that combines 42 simulation tasks with 18 real-world ones.
On the simulation leaderboard, GO-3 Nano Preview posts an average score of 61.90 and an average success rate of 56.42 percent. The benchmark's own update log records the entry as added on October 27 and contributed by AgiBot, on the agent track. Second place is PyRUA-Lean from the PKU DAGroup team at 55.26 and 50.35 percent, so the margin at the top is 6.64 points.
RoboDojo breaks the simulation score into six dimensions, and GO-3 Nano Preview does not lead all of them. It is first on precision manipulation with 78.65, on long-horizon execution with 68.21 and on open-ended semantic tasks with 61.71. On memory it scores 64.50, behind HexaShell-Preview at 84.53 and EmbodiedEvo at 79.40. On standard generalisation it scores 45.93, behind RobotEra's VPP2 at 52.14, and on randomised generalisation it scores 26.97, well behind entries that specialise there.
The benchmark classifies the model as an agent plus world-action-model system rather than a plain vision-language-action policy, which fits the direction AgiBot has been taking since GO-1 and its GE-Act world-action work. Chinese coverage of the result quoted the same headline numbers, with some outlets presenting the per-dimension scores as uniformly first; the leaderboard itself is the stricter record, and that is what is used here.
The entry sits in a crowded field. RoboDojo's simulation board now lists dozens of policies from Chinese and international teams, including Shanghai AI Lab, RobotEra, Galaxea AI, Horizon Robotics, Tencent Robotics X, Li Auto and Xiaomi Robotics, alongside reference entries run by the benchmark team itself. Separately, a system called RoboDoctor, from a team led by a Tsinghua University professor, reported 57.10 on RoboDojo's 54-task simulation summary this week, giving some sense of how quickly the numbers at the top are moving.
AgiBot has not published a technical report for GO-3 Nano Preview, and the leaderboard entry is marked as a preview, so the figures describe a work in progress rather than a shipped model. What the result does show is that the company's model line, rather than only its robot hardware, is now being measured against everyone else's on a shared board.
SEE ALSO: AgiBot G2 Max Goes to Work in JD Logistics' Robotic Warehouse · AgiBot Unveils GE-Act 2.0 Native World-Action Model With 100x Data Scaling