Reportedly: AgiBot Unveils GO-3 Preview Robot Model, Tops Three Leaderboards and Plans Open Source
Reported by Pandaily and IT Home · not yet confirmed by the company or a second independent outlet. We update this page when it is.
AgiBot released GO-3 Preview, an embodied model using a WRAM architecture that tops RoboDojo, RoboTwin and RoboColiseum, and plans to fully open-source it by end of 2026, Pandaily reported.
China context
- Original name
- 智元机器人
- Outside China
- Not stated in the sources yet
- Claims
- Company-reported; not yet independently evaluated
- For builders
- Developers can prepare for the open-source release by studying the WRAM architecture and planning integration with existing robot stacks.
- For investors
- AgiBot's open-source move may pressure commercial embodied AI vendors and create opportunities for startups building on GO-3.
Translated from Chinese. Quotes and facts link to the original sources.
AgiBot's GO-3 Preview is an embodied model built on a WRAM (World Reasoning Action Model) architecture that connects task reasoning, future prediction and action generation. It has topped the RoboDojo, RoboTwin and RoboColiseum leaderboards and is scheduled to be fully open-sourced by the end of 2026. The model supports zero-shot and few-shot task completion in home, kitchen, multi-robot collaboration and dexterous manipulation scenarios without additional fine-tuning.
The WRAM architecture integrates task reasoning, future prediction and action generation into a single model, enabling robots to interpret instructions, anticipate outcomes and produce executable actions. The model is trained on heterogeneous data including human first-person videos and robot trajectories, and demonstrates zero-shot and few-shot capabilities across diverse manipulation tasks.
Robot developers outside China will gain access to a competitive open-source embodied model by end of 2026, reducing the cost of building general-purpose manipulation systems and intensifying pressure on Western robotics labs to match its benchmark performance.
Open-sourcing GO-3 Preview could lower entry barriers for robotics startups and enterprises seeking advanced embodied AI capabilities, while positioning AgiBot as a key supplier of foundational robot models.
The next verifiable signal is the actual open-source release of GO-3 Preview weights and code by the end of 2026, along with independent replication of its leaderboard results.