- GO-2 aims to solve the challenge of disconnected robot planning and execution through a pioneering action chain-of-thought.
- The release is a key component of Agibot's "AI Launch Week" series of events held this week.
Chinese humanoid robot startup Agibot has unveiled its next-generation embodied foundation model, GO-2, marking a new step in the company's push toward practical robotics.
Fully named Genie Operator-2, the model's core objective is to bridge the long-standing semantic-motor gap in the field of embodied intelligence and achieve the "unity of knowledge and action," according to a company announcement on Thursday.
In the past, robotic systems often faced the dilemma of disconnected high-level semantic planning and low-level action execution. While robots could generate perfect task plans, they struggled to complete them stably in real, complex environments.
Compared to its predecessor, GO-1, GO-2 focuses on solving the challenge where robots "think clearly but act unreliably," bridging the last mile from planning to execution, Agibot said.
The model pioneers an action chain-of-thought technology, allowing robots to reason and plan directly within the action space.
Meanwhile, GO-2 introduces an asynchronous dual-system architecture that closely integrates low-frequency planning with high-frequency execution. This enables robots to "think before they act" just like humans, significantly reducing deviations during actual execution.
Across multiple industry benchmarks, GO-2's performance has comprehensively shattered existing state-of-the-art records. Its task success rate and generalization capabilities surpass current mainstream models such as π0.5 and Nvidia's GR00T, according to the company.
The release is a key component of Agibot's "AI Launch Week" series taking place this week. The company plans to announce a core technological breakthrough in the field of physical AI every day from April 7 to April 14.
On Tuesday, Agibot open-sourced its embodied dataset, Agibot World 2026, and rolled- out the Genie Sim 3.0 simulation platform on Wednesday. These initiatives are designed to address the data scarcity problem facing the industry and accelerate commercial deployment.
Notably, Agibot celebrated the official roll-out of its 10,000th humanoid robot off the production line last week. This mass production speed has exceeded market expectations, putting the company ahead of rival Tesla in the pace of commercialization.
Company executives said that 2026 will be a critical stage for general-purpose embodied robots to accelerate into large-scale commercial deployment.
