Wednesday Aug 26
SPORTOther

Robots Beat Bolt, Then Lost To Beans

26AUG
8.86 SECONE BEAN

A Chinese humanoid ran 100 meters in 8.86 seconds this week. The same machines struggle to pick up soybeans with tweezers. Beijing's robot games close today.

Tiangong Ultra clocked 9.39 seconds on August 22, beating Usain Bolt's 9.58. Three days later it ran 8.86 in the semifinals and broke its own record.

Eight of the 51 events test hands. Opening lids. Connecting cables. Picking soybeans with tweezers. Those are the ones robots lose. Legs are solved. Fingers are not.

666 teams brought over 2,000 robots to Beijing. The sprint records make the headlines. The dexterity events tell you what a home robot still cannot do.

full brief & sources

⚡ Why this matters

  • Locomotion is close to solved and manipulation is not. The gap decides which jobs a humanoid can actually take.
  • The sprint numbers are what gets quoted in funding decks. The bean event is what a warehouse buyer should watch.
  • China is running the only large public benchmark for humanoid hardware, which sets the reference points everyone else argues against.

🔍 What happened

  • The 2nd World Humanoid Robot Games ran in Beijing through August 26, 2026.
  • 666 teams entered more than 2,000 humanoid robots across 51 events.
  • Tiangong Ultra ran the 100 metres in 9.39 seconds on August 22, under Usain Bolt's 9.58-second human record.
  • It then clocked 8.86 seconds in the semifinals, breaking its own mark.
  • Eight scenario-based events test dexterous hands: opening lids, assembling power tools, unpacking boxes, connecting cables, picking soybeans with tweezers.
  • The soybean-and-tweezers contest is designed to test precision and tool handling for home and workplace use.

💬 Smart takes

  • Gizmodo: framed the week as robots beating Bolt and then meeting their match in beans.
  • Forbes (John Koetsier): read the event list itself as a market map for where humanoid capability is heading.
  • Skeptic: a purpose-built sprinter on a prepared track is a demo, not a product. Nobody is buying a robot to run 100 metres.

🧭 Where this goes

  1. Likelysprint records keep falling and stop being news within a year.
  2. Likelydexterity scores become the metric that investors and buyers actually track.
  3. Possiblea Western equivalent event launches to avoid ceding the benchmark to Beijing.
  4. Wild Carda dexterity result jumps far enough that home-robot timelines get pulled forward.

🥄 The Spoon Take

Read the event list, not the headline. Running fast is a controls problem on a known surface and it is basically finished. Picking up a soybean with tweezers is perception, force control and recovery from failure, and it is not. Any humanoid pitch that leads with speed and skips the hands is selling you the easy half.

🤔 Pushback

This is a showcase, not a controlled benchmark. Teams tune for the event and the times are not independently verified. Reading it as a capability measurement is a stretch.

·Gizmodo - Bolt, then beans·CGTN - 8.86 second run·Al Jazeera - sub-9-second 100m·CGTN - bean-picking challenge