AnswersPublished15 sources
AgiBot’s GO-3 Nano Leads RoboDojo’s Simulation Board—but Not Every Metric
GO 3 Nano Preview topped RoboDojo’s simulation agent track with a 61.90 average score and 56.42% success rate. RoboDojo’s leaderboard lists a 6.64 point gap over second place, PyRUA Lean.[11][17] Published dimension scores differ between Chinese language reports and the leaderboard snapshot, so claims of category wi...
Published byEdited with GPT-6 LunaImages generated with GPT Image 2
Research answer
Key takeaways
- GO 3 Nano Preview topped RoboDojo’s simulation agent track with a 61.90 average score and 56.42% success rate.
- RoboDojo’s leaderboard lists a 6.64 point gap over second place, PyRUA Lean.[11][17]
- Published dimension scores differ between Chinese language reports and the leaderboard snapshot, so claims of category wins need that caveat.[8][10][12]
AI Prompt
openai.comCreate a landscape editorial hero image for this Studio Global article: What does AgiBot’s GO-3 Nano Preview’s October 27, 2026 debut atop RoboDojo’s simulation leaderboard reveal about its performance and signif. Article summary: GO-3 Nano Preview’s debut indicates a strong overall simulation result, not proof of uniformly superior robotic intelligence: it topped RoboDojo with a 61.90 average score and 56.42% success rate.[2] Its broader signific. Topic tags: general, general web, user generated. Style: premium digital editorial illustration, source-backed research mood, clean composition, high detail, modern web publication hero. Use reference image context only for broad subject, composition, and topical grounding; do not copy the exact image. Avoid: logos, brand marks, copyrighted characters, real person likenesses, fake screenshots, UI text, readable text, watermarks, charts with fa
AgiBot’s GO-3 Nano Preview took first place on RoboDojo’s simulation agent track with a 61.90 average score and a 56.42% success rate. The result is a meaningful benchmark lead, but it applies to simulation—not to RoboDojo’s real-world tasks—and does not show that the model is strongest in every capability.12
17
What the leaderboard result measures
RoboDojo describes a benchmark spanning 42 simulation tasks and 18 real-world tasks. Its October 27 update lists GO-3 Nano Preview as an AgiBot-contributed entry on the simulation leaderboard’s agent track.17 The 61.90 score and 56.42% success rate are therefore simulation results; they should not be presented as results across all 60 tasks or as a demonstration of real-robot performance.
12
17
The leaderboard places the preview ahead of PyRUA-Lean, listed at 55.26 with a 50.35% success rate. That makes the score gap 6.64 points and the success-rate gap 6.07 percentage points.11
17 These figures describe performance on this benchmark, not a general measure of robotic capability in every setting.
A strong aggregate score, not a clean sweep
RoboDojo’s published leaderboard snapshot gives GO-3 Nano Preview a generalization score of 33.66, precision score of 76.00, long-horizon score of 53.50, memory score of 63.67 and open-task score of 55.25.12 The same snapshot shows HexaShell-Preview scoring higher on generalization and memory, at 34.50 and 81.00 respectively.
12 So the overall first-place ranking does not mean GO-3 leads every listed dimension.
There is also a discrepancy in reported dimension scores. Chinese-language coverage gives GO-3 Nano Preview scores of 78.65 for precision, 68.21 for long-horizon execution and 61.71 for open-ended semantic tasks, describing all three as first-place results.8
10 Those numbers do not match the dimension scores in the leaderboard snapshot.
12 Without a clear explanation of the difference, it is safer to distinguish the media-reported figures from the benchmark’s displayed row rather than combine them or treat them as interchangeable.
Why the agent-track entry matters
The benchmark identifies GO-3 Nano Preview as an agent-track entry contributed by AgiBot.17 That makes the result relevant not only as a robotics-company milestone but also as an evaluation of an AgiBot system on a shared manipulation benchmark. It offers a point of comparison with other listed policies, while remaining specific to RoboDojo’s tasks, scoring and simulation setup.
12
17
The entry does not, by itself, establish how the system is built or how much of its performance comes from any particular model architecture. Nor does a simulation ranking establish deployment reliability. The available leaderboard information supports the performance claim; it does not settle those broader technical questions.
The bottom line
GO-3 Nano Preview’s debut is best read as a strong aggregate simulation result: first on RoboDojo’s agent track, with a 61.90 average score and 56.42% success rate. The result is notable, but it is narrower than a claim of universal robot intelligence: RoboDojo also includes real-world tasks, the system does not lead every dimension in the published snapshot, and reported category scores do not fully align.8
12
17
Studio Global AI
Continue your research
This page includes a source-backed answer you can continue inside Studio Global.
People also ask
What is the short answer to "AgiBot’s GO-3 Nano Leads RoboDojo’s Simulation Board—but Not Every Metric"?
GO 3 Nano Preview topped RoboDojo’s simulation agent track with a 61.90 average score and 56.42% success rate.
What are the key points to validate first?
GO 3 Nano Preview topped RoboDojo’s simulation agent track with a 61.90 average score and 56.42% success rate. RoboDojo’s leaderboard lists a 6.64 point gap over second place, PyRUA Lean.[11][17]
What should I do next in practice?
Published dimension scores differ between Chinese language reports and the leaderboard snapshot, so claims of category wins need that caveat.[8][10][12]
Sources
robodojo-benchmark.comRoboDojo: A Unified Sim-and-Real Benchmark for Comprehensive Evaluation of Generalist Robot Manipulation Policies
pandaily.comAgiBot's GO-3 Nano Preview Tops RoboDojo's Simulation Leaderboard With a 61.90 Average Score
hk.finance.yahoo.com智元GO-3 Nano Preview在RoboDojo榜超越GPT-6 Astra