AGIBOT GO-3 Nano Tops RoboDojo; Unitree H2-X Claims 8-Hour Battery

On October 27, AGIBOT’s GO-3 Nano preview scored 61.90 average and 56.42% success on RoboDojo, ahead of GPT-6 Astra; Unitree’s H2-X demonstrated kicking open a car door and lifting a vehicle, with a claimed 8-hour battery. The two data points mark the frontier in decision intelligence and in physical execution and endurance.

Toyota has disclosed a target of 400,000 humanoid robot deployments. Shanghai’s intelligent manufacturing plan focuses on core technologies and foundational components, while China’s eighth batch of new occupations includes an embodied AI robot application technician. Policy and volume targets now converge on components, deployment and maintenance, making supply-chain readiness the precondition for scale. Demo performance alone is insufficient.

Leaderobot counts nearly 190 seed, angel and early-stage deals, indicating crowded capital at the startup end. MoSense raised about RMB100 million in an angel-plus round led by CDH Investments and Oasis Capital, with LimX Dynamics and Loongson Venture Capital participating; electronic skin is a core component. By this measure, early-stage funding density coexists with engineering deployment bottlenecks, and Unitree’s more than 65% rebound in seven trading days remains a thematic signal that requires order and production-line validation.

RoboParty open-sourced a full-stack humanoid, including mechanical drawings, BOM, motion algorithms and a troubleshooting database, lowering entry barriers. Open-source diffusion may compress prototype integration premiums and shift value toward high-reliability components and scenario data. Two observation points: whether benchmark scores convert into real-world success rates, and whether component and maintenance occupation standards enter procurement and training systems.