AGIBOT’s GO-3 Nano Preview Tops RoboDojo Benchmark, Outperforming GPT-6 Astra

RoboDojo, an evaluation benchmark for embodied intelligence, published its latest leaderboard on October 27. AGIBOT’s GO-3 Nano Preview ranked first with an average score of 61.90 and an average success rate of 56.42%, surpassing GPT-6 Astra.

The overall result was accompanied by first-place finishes in three subcategories. GO-3 Nano Preview scored 78.65 in fine manipulation, 68.21 in long-horizon execution, and 61.71 in open semantic understanding. Those dimensions test precision control, extended multi-step task completion, and language-grounded scene understanding.

RoboDojo compares embodied models through standardized evaluations and reports average scores and success rates. The latest ranking therefore offers a comparative view of model performance across both aggregate and task-specific measures.

The leaderboard release follows growing interest in embodied AI, where models are assessed not only on language tasks but also on their ability to act in physical or simulated environments. Success rates and category scores are intended to capture reliability as well as capability.

The outcome places GO-3 Nano Preview ahead of GPT-6 Astra on the current leaderboard and highlights progress in systems that combine perception, planning, and control. The results were reported by Wallstreetcn’s Express desk.