A 98% task success rate is the headline number from a new humanoid robot developed in China, which its developers say can handle electric vehicle assembly work and general factory operations.

The claim positions the machine alongside a growing field of Chinese humanoids — from Unitree, UBTech and others — that OEMs and tier suppliers are increasingly trialing on production lines. As with every vendor announcement in this space, the 98% figure is a developer-reported benchmark, not an independently verified production metric. Buyers should read it as a lab or pilot result until it shows up in confirmed line-rate throughput data from a named plant.

According to the report from Interesting Engineering, the robot can perform tasks relevant to EV assembly and broader factory operations. The machine joins a wave of Chinese humanoid platforms now being tested in automotive settings, where automakers are under pressure to automate final assembly, material handling and quality inspection — work that traditional fixed automation handles poorly because of high variant counts and frequent model changeovers.

The success-rate claim matters because task reliability, not raw capability, has been the gatekeeping metric for humanoids on vehicle lines. A robot that completes 98 of 100 pick-place or fastening attempts moves from demonstration to deployable. Whether this particular platform has crossed that threshold on an actual EV line, at takt, with production tooling, is the open question.

Chinese OEMs and their tier-one suppliers have moved faster than most Western counterparts in trialing humanoids. The test cases so far concentrate on repetitive mid-line tasks: component transfer, kitting, and inspection stations where cycle-time penalties of a failed attempt stay bounded. Final assembly and battery-line work remain harder trials.

For plant operators, the practical questions are the standard ones: payload, cycle time, mean time between failures, integration cost against a cobot alternative, and whether the vendor can support the machine at scale. A 98% success rate on a controlled task set does not yet answer any of them.

What to watch next: whether the developer names an automotive customer or pilot plant, and whether the platform moves from reported benchmarks to published throughput data on a working line. A confirmed deployment at a named EV factory would be the signal that separates this announcement from the crowd.