In One Year, Robots Went from a "Marathon Joke" to Surpassing the "Strongest Humans"

Robots outrun humans — reportedly

It has been reported that humanoid robots at the 2026 Beijing Yizhuang Humanoid Robot Half Marathon have overtaken the best human times. Honor (荣耀)’s flagship humanoid "Lightning" reportedly won the race with a net time of 50:26, beating last year’s robot champion, Tiangong Ultra (天工Ultra), which finished in 2:40:42 in 2025. For perspective: the current human men’s half‑marathon world record is 57:20 (Kiprimo, March 2026). Surpassing the “strongest humans” in pure speed — in just one year — is a striking milestone. How did this happen so fast?

From a fiasco to a field of serious players

A year ago the sport looked like a public embarrassment: at the inaugural 2025 race only six of 20 teams finished, with robots veering off course, losing power at the start line, or needing engineers to run alongside and physically intervene. This year, it has been reported that the event swelled to some 300 robots from more than 100 teams — including research institutes, university squads, and new entrants such as Honor (荣耀) and Amap (高德). Beijing Humanoid Robot Innovation Center (北京人形机器人创新中心)’s autonomous Tiangong Ultra completed the half marathon in about 1:15 with zero human intervention, and Unitree (宇树科技)’s H1 showed dramatic improvements in qualifying runs. The top three finishers were all autonomous machines from Honor’s teams, and the race rules now explicitly reward autonomy (remote‑controlled times are penalized by a 1.2 multiplier), reflecting a deliberate shift from remote‑control demos toward true embodied intelligence.

Money, data and the hard problems ahead

The speed gains are real — but so are the limits. It has been reported that 2025 investment into China’s robotics sector ballooned to between RMB 511 billion and RMB 735.43 billion, and that financing has continued strongly into 2026. One company reportedly closed a record Pre‑A of more than $450 million led by Hillhouse and Sequoia China. Yet leading researchers and company filings warn that motor control and locomotion are only one axis of capability. Experts say large gaps remain in task‑level, multimodal data for training embodied models — a shortfall of perhaps two orders of magnitude, according to Fudan professor Xiao Yanghua — and that dexterous manipulation, hardware standardization and affordable data collection remain unsolved. Unitree’s IPO filing, for example, earmarks a sizable share of proceeds for large‑scale data and model efforts, underscoring how central data is to the next phase.

Will speed translate into useful, widely deployed robots? It has been reported that some founders expect shipments to surge into the hundreds of thousands if a major algorithmic breakthrough arrives, but analysts caution that geopolitical constraints on advanced chips and AI compute — Western export controls and trade policy around high‑end semiconductors — could complicate timelines. For now the headline is simple: machines that were a running joke a year ago can now outrun elite humans on a stopwatch. The harder question is whether they can leave the track and reliably handle the messy, multi‑tasked world beyond it.