Tiangong Ultra Runs 100 m in 8.64 s

On August 26, 2026, Tiangong Ultra completed the 100 meters in 8.64 seconds at the second World Humanoid Robot Games in Beijing.

It was not just the fastest 100-meter time recorded by a humanoid robot at the Games. It was also 0.94 seconds faster than Usain Bolt’s 9.58-second human world record, set in Berlin 2009.

The result came at the end of an extraordinary five-day progression. Tiangong Ultra first ran 9.39 seconds during the opening-day preliminary round. Three days later, it improved to 8.86 seconds in the large-group semifinal. In the final on August 26, it lowered the mark again to 8.64 seconds — and won the large-group 100-meter title.

The numbers were spectacular.

The finish, however, told another story.

Together, the record and the chaotic finishes offered a revealing snapshot of humanoid robotics in 2026:

Robots are becoming dramatically faster — but controlling that speed remains a different problem.

Tiangong Ultra’s 8.64-second final was not an isolated breakthrough. The same robot repeatedly lowered the 100-meter record during the Games.

The progression was remarkable: the robot cut 0.75 seconds from its opening mark in only four days. The final was also 0.22 seconds faster than its semifinal.

The 2025 World Humanoid Robot Games provide an even more striking comparison. Tiangong Ultra won the inaugural 100-meter event in 21.50 seconds. One year later, the winning time had fallen to 8.64 seconds.

Tiangong Ultra was developed by the Beijing Humanoid Robot Innovation Center, also known as X-Humanoid （北京人形机器人创新中心 / X-Humanoid）.

According to the development team, the latest version received upgrades in several areas, including its joint motors, mechanical configuration, thermal management and control system.

• Joint motors upgraded for higher torque and rotational speed — explosive acceleration.
• Lightweight + aerodynamic body changes to reduce aerodynamic drag during sprint.
• Body-control capability upgraded — specifically acceleration training for the 100 m event.
• Thermal management strengthened to survive full-throttle sprints + three rounds in five days.

The 8.64-second performance was not simply the result of installing a more powerful motor. It was the result of coordinating mechanical power, lightweight design, thermal management and high-speed motion control into a system capable of maintaining balance while moving at extreme speed.

A humanoid robot does not have to run exactly like a human. In fact, at these speeds, its mechanical design creates different possibilities.

Reports from the event noted that Tiangong Ultra takes more steps over 100 meters than an elite human sprinter, while using a much faster cadence. Its movement is optimized around the characteristics of its motors, joints, body structure and control system rather than the biomechanics of a human athlete.

This is one reason why robot athletics can be useful as an engineering benchmark: developers can optimize a machine for measurable physical objectives rather than forcing it to reproduce human movement exactly.

The comparison with Usain Bolt is undeniably striking.

Bolt’s 9.58-second performance remains the official men’s 100-meter world record recognized by World Athletics. Tiangong Ultra’s 8.64-second time is faster by the stopwatch.

But the two performances should not be treated as equivalent athletic records.

Humans and robots compete under fundamentally different physical and regulatory conditions. The humanoid competition uses its own starting procedures, track setup and rules, and the result is not a World Athletics record.

The more meaningful comparison is therefore not “Is Tiangong Ultra a better sprinter than Usain Bolt?” — it is “How quickly can a bipedal machine now accelerate, maintain dynamic balance and cover 100 meters under controlled conditions?” On that question, the answer is remarkable.

The most revealing images from the 100-meter events were not necessarily the fastest ones. They were the ones immediately after the finish.

The contrast was striking. The robots had become good enough to run at speeds beyond the human world record — but some still struggled with what should happen after the race.

This reveals a fundamental robotics problem: high-speed locomotion and controlled deceleration are two different engineering challenges.

A robot can be optimized to accelerate aggressively and maintain a fast gait without necessarily having an equally sophisticated strategy for slowing down, changing direction or recovering from unexpected disturbances.

• On a straight athletics track → padded barriers absorb the stop.
• In a warehouse / factory / home → no guaranteed crash barrier at the end of movement.

The World Humanoid Robot Games were designed to test much more than running speed. The 2026 event brought together 666 teams and more than 2,000 robots from 16 countries, competing across 51 events. Alongside athletics, the Games included football, combat sports and scenario-based tasks designed to test practical capabilities.

Those practical environments represent a much harder test of embodied intelligence. A robot in a factory may need to:

• Start and stop repeatedly — not accelerate once in a straight line.
• Navigate around people and equipment — crowds, pedestrians, moving forklifts.
• Turn in tight spaces, carry objects while walking, recover after losing balance.
• Adapt to unexpected obstacles (wet floor, dropped package).
• Operate for hours rather than seconds — energy + thermal endurance.

Despite these limitations, the achievement should not be dismissed.

Running a humanoid robot at 8.64 seconds over 100 meters requires extremely fast and coordinated control of multiple joints while maintaining a dynamically unstable body. The robot must continuously manage:

• Center of mass / moment balance through every step.
• Foot placement accuracy at sprint cadence.
• Actuator output per-joint: torque + current + phase.
• Body orientation — roll / pitch / yaw under high vibration.
• A failure in any one of these systems can result in a fall.

The fact that Tiangong Ultra could repeatedly improve its performance during a single competition suggests that the underlying hardware and control stack have reached a new level of maturity.

More importantly, locomotion is not an isolated capability. Walking, running, climbing stairs and navigating uneven environments all depend on the same broader problem:

“How does a machine move its body reliably through the physical world?”

The Tiangong team itself describes running as an important expression of humanoid robots’ generalized mobility — the ability to reach places that humans can reach. That makes the 100-meter race more than a spectacle. It is a highly visible benchmark for embodied mobility.

Tiangong Ultra’s 8.64-second run represents one of the clearest demonstrations yet of how rapidly humanoid locomotion is advancing. In just one year, the winning 100-meter time at the World Humanoid Robot Games dropped from 21.50 seconds to 8.64 seconds.

But the dramatic finishes also provide an equally important lesson.

Being able to run fast is not the same as being able to move reliably.

The next major challenge is not simply to make humanoid robots faster. It is to make them capable of controlling that speed: to accelerate, brake, turn, recover and continue working — without needing a crash barrier or a team of people waiting with stretchers.

That is the difference between a robot that can win a race, and a robot that can actually do a job.