Science1 distinct publisher3 min readPublished
Tiangong Ultra beat four men's world records in robot-only categories at the Beijing games, but the only mark with a clean comparator is a standing high jump three times higher than last year's robot best.
The Scientist · Science desk
Compiled by The ScientistSomething wrong?How this is made
Convert the marks into speeds and the informative comparison is between the robot's own three runs, not against the men in the record books. The winning 100 m time averages 11.57 metres per second [1]. The 1,500 m, 141.6 seconds of it, averages 10.59 [2], which is 92 percent of the sprint pace [3]. Among the human records that ratio is closer to 70 percent, with Usain Bolt's 9.58 seconds in Berlin at 10.44 m/s and Hicham El Guerrouj's 3:26.00 in Rome at 7.28 m/s [7][11][4][5][6]. Stranger still is the 400 m. At 38.15 seconds it averages 10.48 m/s, about a tenth of a metre per second slower than the same robot's 1,500 m pace [8][7].
That ordering does not come out of a tiring athlete, and it means the three times are not three points on one endurance curve. Something other than stamina set each of them, and the material does not say what: gearing, control margins, battery and thermal headroom, or how hard each final was actually contested. The sprint progression inside the meet points the same way, since 0.75 seconds, or 8 percent, came off the 100 m time in a matter of days [8]. That is what tuning looks like on a timing sheet.
Scott Walter, Robotics Research Diligence director at the robotics investment fund RoboStrategy, told Live Science the marks came from "mostly mechanical optimization to perform a single task extremely well" under a "The Best Part is No Part" principle, and that "locomotion is mostly solved as witnessed by these games, albeit with optimizations" [18][21][19]. The mechanism he describes is mass placement: take degrees of freedom out of the arms, hips and ankles, and the machine loses weight while its remaining distal mass sits higher up the kinematic chain, which improves movement efficiency [20].
The high jump is where the year comparison is clean and the human comparison is not. Measured against the best robot mark at the 2025 inaugural games, this year's standing clearance is a factor of three on an identical event definition [9], the only figure in the set with a real baseline behind it. Measured against Javier Sotomayor's 2.45 m in Salamanca in 1993, it is a different event, because he used a curved approach, converted horizontal speed through a one-foot takeoff and cleared the bar with a Fosbury flop, none of which a two-footed standing jump can draw on [15][16].
Distance running and high standing jumps demand powerful actuators, fast balance corrections, foot-placement control and software that compensates for small errors in real time, per Live Science's account [17], and at least three robots finishing the 1,500 m inside El Guerrouj's mark suggests more than one team now has all four working together [10]. What the timing sheet does not measure is what happens when the wrists, the hands and a payload go back onto the machine.
Ranked by verification strength, evidence, and original report placement.
The five-day 2026 World Humanoid Robot Games were held between Aug. 22 and 26 at Beijing's National Speed Skating Oval and drew 2,056 robots from 666 teams across 16 countries.
Events at the games included football and martial arts as well as practical applications such as warehouse logistics, factory-style assembly and charging tasks.
Humanoid robots at the games beat men's athletics world records in the 100 m, 400 m, 1,500 m and high jump, but competed under robot-specific categories and conditions, so the performances are not official replacements for the human records.
Tiangong Ultra, a humanoid robot developed by the Beijing Humanoid Robot Innovation Center, won the large-size 100 m final in 8.64 seconds on Aug. 26.
Tiangong Ultra set an early 100 m mark of 9.39 seconds, improved to 8.86 seconds in qualifying, then ran 8.64 seconds in the final.
Usain Bolt ran 9.58 seconds over 100 m at the 2009 World Athletics Championships in Berlin.
Distinct publishers with included, body-backed reporting in this cluster.
Follow any of these and your For You feed starts watching them — no settings page required.
invest
Beijing fielded 2,000 humanoids the same month Washington banned importing them1 distinct publisher
science
Hammering nails exposed the gripper limit that Beijing's 8.64-second sprint hid2 distinct publishers
leadership
China's robot games fielded 2,000 humanoids from 660 mostly domestic builders1 distinct publisher
product
A humanoid beat Bolt's 100m mark on a course that removed everything hard about the job5 distinct publishers
Evidence-backed comparisons of source perspectives and observed adoption signals. Read the methodology
Which Builder, Operator, and Investor concerns the observed source mix emphasized—not a truth score.
Evidence, demonstrated adoption, hype gap, incentives, and confidence are assessed independently, each on its own current evidence. How these are measured.
One outlet, one named expert
Every robot time and distance here traces to Live Science, which names no results sheet, organiser statement or timing method behind them. The human comparators are checkable against athletics records; the machine marks are not checkable at all from what we hold. What raises the floor is that the same report carries its own limits — robot-only categories, a standing jump standing in for a running one — and one named engineer on the record with specifics rather than adjectives.
Five days inside one oval
What has actually been observed is a competition: 2,056 machines, 666 teams, sixteen countries, and events that simulate logistics, assembly and cable-plugging rather than perform them for a customer. The scale of entry is real and it is bigger than last year's field. Beyond the stadium, this reporting does not establish any deployment or buyer, nor does it say anything about working hours -- and by Walter's account the record holders could not walk to the podium.
Record framing runs ahead of the machines
The headline asks whether a robot is faster than Usain Bolt, and the table beneath it prints four human records with robot times below them. The qualifications are all in the same piece: robot-only categories, straight-line acceleration with no ability to run the curve or stop cleanly, no hands for a relay, and a 2.88 m standing jump measured against a running high jump with different mechanics. The one mark with an honest predecessor is that jump, up from 0.95 m at last year's games. Live Science is not hiding the gap; the presentation still invites the reading the caveats forbid.
Sole analyst sells the sector
The story's only interpretive voice works for a fund that sells robotics and physical-AI exposure, which Live Science says plainly in the attribution line. Walter's quotes run against that interest — 'barely humanoid', unable to stop, no hands — so the disclosure matters less than it might. Pull the other way and the record times themselves reach us only through the parties who set them, with the developers named and the failures counted by nobody.
Fair on the mechanics, thin on the numbers
The mechanics hold up better than the numbers: a named engineer explains what these machines were built to do and why the comparisons to human records don't hold, and the reporting itself concedes the category problem, so that much is fairly solid. The times are the weak point, since not a single robot time here is independently verified, and if any of them is off, the pace arithmetic that makes this story interesting moves with it.