Back to stories
Research

Honor's Autonomous Humanoid Robot Wins Beijing Half-Marathon in 50:26, Outpacing Human World Record

Michael Ouroumis2 min read
Honor's Autonomous Humanoid Robot Wins Beijing Half-Marathon in 50:26, Outpacing Human World Record

A humanoid robot running autonomously for Chinese smartphone maker Honor won Beijing's second annual humanoid robot half-marathon on Sunday, finishing the 21-kilometer course in 50 minutes and 26 seconds. The time was fast enough to beat the current men's human world record of 57:20, held by Ugandan runner Jacob Kiplimo, and marked one of the clearest demonstrations yet that legged robotics is closing — and in some respects overtaking — human performance on long, unstructured courses.

The race took place in Beijing E-Town, the capital's economic-technological development zone, which has emerged as a showcase corridor for Chinese physical AI. A separate remotely-controlled Honor robot actually crossed the finish line first in 48 minutes and 19 seconds, but the championship went to the autonomous entrant under the event's weighted scoring rules, which reward robots that navigate without a human pilot in the loop. Runners-up, also from Honor, finished in roughly 51 and 53 minutes.

From 2h40m to sub-51 in a year

Last year's inaugural race was won in 2 hours, 40 minutes and 42 seconds — more than double a typical human half-marathon winner's pace, and a reminder of how recently humanoids struggled to walk outdoors at all. Compressing that into a sub-51 autonomous run in twelve months is the kind of curve that has become familiar in frontier AI: early prototypes that look comically slow, followed by an abrupt jump once training pipelines, actuator designs and perception stacks catch up.

Honor's test development engineer Du Xiaodi said the winning machine was built around legs roughly 95 centimeters long and an in-house liquid-cooling system designed to keep actuators from throttling over the course of the race. Cooling has become one of the quiet bottlenecks for endurance tasks on humanoids, where high-torque motors tend to overheat well before batteries run flat.

112 teams and an international field

Organizers said 112 teams entered, up from about 20 in 2025, including five international teams from Germany, France and Brazil. Roughly 40% of the robots ran with autonomous navigation; the rest were remotely piloted along the course. The jump in entries mirrors the broader rush of capital into humanoid platforms over the past year, with Chinese OEMs, automakers and consumer electronics firms all standing up robotics divisions.

Why the record claim deserves a caveat

The comparison to Kiplimo's 57:20 is eye-catching but not apples-to-apples. The robot course is not a World Athletics-certified route, and the event allows pit stops for battery swaps and adjustments. Still, the symbolic line has been crossed: in at least one public, timed 21-kilometer race, a legged machine running without a remote pilot covered the distance faster than any human ever has. For researchers tracking the real-world generalization of physical AI, that is a data point worth pinning to the wall.

Learn AI for Free — FreeAcademy.ai

Take "AI Essentials: Understanding AI in 2026" — a free course with certificate to master the skills behind this story.

More in Research

Northwestern's Printed Artificial Neurons Talk Back to Living Brain Cells
Research

Northwestern's Printed Artificial Neurons Talk Back to Living Brain Cells

Northwestern engineers have printed soft, flexible artificial neurons that can activate living brain tissue, a Nature Nanotechnology result that points toward a new generation of brain-machine interfaces and brain-like computing hardware.

3 min ago3 min read
Agents of Chaos: New Paper Documents Dozen Dangerous Actions by OpenClaw AI Agents
Research

Agents of Chaos: New Paper Documents Dozen Dangerous Actions by OpenClaw AI Agents

A 20-researcher study titled 'Agents of Chaos' documented roughly a dozen dangerous actions by autonomous AI agents, from deleting email inboxes to leaking medical and financial records — fueling a wider expert warning on April 19 about the cybersecurity risks of the agentic AI boom.

6 hours ago3 min read
Anthropic's Mythos Is Finding Bugs Faster Than Open-Source Teams Can Patch Them
Research

Anthropic's Mythos Is Finding Bugs Faster Than Open-Source Teams Can Patch Them

Bloomberg reporting this week highlights a lopsided new reality: Anthropic's Mythos model has surfaced thousands of high- and critical-severity vulnerabilities across major operating systems and browsers, but fewer than 1% have been patched by maintainers.

1 day ago3 min read