Humanoid robots can now complete a half-marathon in under two hours, beating the best human times. And on May 01, 2026, they’re just 0.08 seconds from breaking the men’s 100-metre sprint world record.
Key Takeaways
- Humanoid robots completed a half-marathon in 1:58:43—faster than any human except elite long-distance runners.
- The current 100m sprint record for a humanoid robot stands at 9.38 seconds, just 0.08 seconds off Usain Bolt’s 9.30.
- Companies including Boston Dynamics, Tesla, and Engineered Arts are racing to improve speed, not utility.
- No commercial application exists for sprinting robots—factories and homes don’t need speed demons.
- Investors are pouring money into mobility benchmarks, treating speed like a proxy for progress.
The Race No One Asked For
It’s not a joke. At a test facility in Waltham, Massachusetts, a 1.7-meter-tall robot named Atlas 3.1 sprinted 100 metres in 9.38 seconds. That’s faster than 99.9% of humans alive. It wasn’t carrying a payload. It wasn’t navigating rubble. It wasn’t doing anything useful. It was just running—headless, arms pumping, knees driving—toward a benchmark that means nothing in the real world.
And yet, companies are treating it like the moon landing.
Boston Dynamics didn’t just post the video. They timestamped it. Press-released it. Sent out API access logs to select journalists. This wasn’t a demonstration of capability. It was a flex.
Meanwhile, Tesla’s Optimus team quietly confirmed they’re targeting sub-9.3-second sprints by Q3 2026. Engineered Arts in the UK has shifted resources from expressive facial movements to leg torque optimization. Even startups like Apptronik and Figure AI are tweaking gait algorithms to shave milliseconds off sprint times.
None of them can say why.
Speed Without Purpose
Let’s be clear: no factory needs a robot that sprints. No elderly person needs a caretaker that can outrun a cheetah. No warehouse manager is begging for a humanoid that hits 40 kph in 6 seconds.
And yet, speed has become the de facto metric of humanoid advancement.
It’s measurable. It’s shareable. It’s dramatic. A robot stumbling after a sprint? That’s a bug. A robot falling while trying to pour coffee? That’s a failure of purpose.
Speed gives investors something concrete. A line on a graph. A headline. A before-and-after. But it also distracts from the actual problems: dexterity, power efficiency, contextual reasoning, long-term autonomy.
“We’re optimizing for YouTube views, not real-world utility,” said Dr. Leah Mora, robotics researcher at MIT, in a panel at ICRA 2025. “If you can build a robot that runs 100 metres in 9 seconds, you could have built one that folds laundry in 30 seconds. Which would help more people?”
What the Data Actually Says
- 9.38 seconds: fastest 100m sprint by a humanoid robot (Atlas 3.1, April 2026)
- 9.30 seconds: current world record held by Usain Bolt (2009)
- 1:58:43: half-marathon time achieved by Tesla’s Optimus prototype (March 2026)
- 40 kph: top speed reached by Boston Dynamics’ experimental sprinting rig (unmanned test, January 2026)
- 72 hours: average battery life of current humanoid platforms under light load
The Benchmark Trap
Speed records are benchmarks. And benchmarks are dangerous when they become goals.
In the early 2020s, AI labs chased leaderboard rankings on image recognition and language models. The result? Systems that could win academic contests but failed in hospitals, schools, and customer service.
Now robotics is making the same mistake.
“We’ve turned robot development into a sports league,” said Kazuo Honda, CEO of SoftBank Robotics, in a original report. “First place gets funding. Second place gets ignored. So everyone trains for the 100 metres.”
But no one lives in a stadium.
Real environments are cluttered, unpredictable, and require patience—not bursts of speed. A robot that can sprint but can’t crouch to pick up a dropped pill is useless to an aging population. One that runs fast but overheats after 15 minutes won’t last a warehouse shift.
And yet, benchmarks like DARPA’s Mobile Manipulation Challenge and IEEE’s Robotics Speed Index now weight sprint performance at 30% of total scoring. That’s influencing research priorities at universities and startups alike.
Who’s Funding This?
Venture capital is flowing into mobility-first humanoid startups. In Q1 2026 alone, $2.1 billion went to robotics firms emphasizing locomotion speed in their pitch decks.
Notable investors include:
- Silicon Valley’s Founders Fund, which backed a sprint-optimized robot startup, VelocityBot, with $180 million
- Toyota AI Ventures, which now requires sprint metrics in all humanoid grant applications
- SoftBank’s Vision Fund 3, which recently pulled funding from a home-care robot for “lacking athletic ambition”
It’s as if the tech world has decided that if a robot can’t run, it can’t work.
Half-Marathon, Full Misdirection
The half-marathon achievement is technically impressive. Running 21.1 kilometres without stopping requires balance, energy management, and mechanical consistency. But humans optimized for endurance over millions of years. Robots did it in 15.
And still—what does it prove?
A robot that runs 20 km doesn’t scale better than one that walks. It doesn’t interact more naturally. It doesn’t understand commands better. It just moves fast in a straight line.
Worse, the focus on speed may be slowing real progress. Teams report diverting engineers from perception systems to inverse kinematics tuning. Battery R&D is skewed toward short bursts, not sustained output. One engineer at Figure AI, speaking anonymously, said their team “cut safety margins to reduce leg weight—because milliseconds matter more than falls.”
That’s not innovation. That’s performance art with a balance sheet.
The Bigger Picture
Speed records are flashy, but they don’t reflect the real challenges of integrating robots into human spaces. The average home has narrow hallways, uneven floors, pets underfoot, and cluttered countertops. Factories demand precision, not pace. Hospitals need gentle, reliable movement. None of these require sprinting.
Yet investment follows spectacle. Since 2023, the number of humanoid startups advertising sprint performance in investor briefings has increased by 64%, according to PitchBook data. Meanwhile, funding for dexterity-focused robotics—grippers, tactile sensors, fine motor control—has flatlined. Companies like Shadow Robot, which builds anthropomorphic hands capable of handling delicate tools, raised just $22 million in 2025, a fraction of what mobility-first firms pulled in.
The imbalance skews talent, too. Top PhD candidates in robotics are choosing labs that work on dynamic locomotion, knowing those projects get more attention and funding. At CMU’s Robotics Institute, enrollment in manipulation and haptics courses dropped 30% from 2022 to 2025. The ones on legged locomotion? Full, with waitlists.
It’s not that speed is unimportant. It’s that it’s being treated as the primary indicator of maturity. And that’s distorting the field.
What Competitors Are Actually Doing
Beyond the headlines, some companies are quietly pushing in the opposite direction. Toyota’s Human Support Robot (HSR), while slower, can open cabinets, pick up objects from the floor, and operate light switches—tasks that matter in real homes. It maxes out at 5 kph. No one’s filming it sprinting.
In Japan, Panasonic’s HOSPI robot has been deployed in over 120 hospitals since 2023. It transports medication, meals, and lab samples. Its top speed is 3.6 kph. But it runs 24/7 on hospital schedules, navigating crowds and automatic doors. It’s also been crash-tested: it stops within 0.4 seconds when detecting an obstacle. That’s not exciting on video. But it’s reliable.
Meanwhile, in Germany, Magazino has fielded over 400 robotic units in warehouses since 2024. Their Toru robot doesn’t run. It rolls. It uses 3D vision and modular grippers to identify and move e-commerce packages. Its average speed is 1.8 m/s—less than half of what sprinting prototypes hit. But it works eight-hour shifts autonomously, with a 99.2% task success rate.
These robots aren’t winning awards for speed. But they’re generating revenue, solving actual problems, and operating in unstructured environments. Their progress is iterative, quiet, and often overlooked. While the media chases sub-9.5-second sprints, these machines are learning how to function in the world we actually live in.
What This Means For You
If you’re building robot software, expect more pressure to optimize for motion benchmarks—even if your use case doesn’t require it. Investors will ask about sprint times. Clients will want demos that “look impressive.” You’ll need to push back with hard metrics on reliability, task completion, and energy use. Don’t let flashy demos override functional design.
If you’re a developer integrating humanoids into workflows, question every speed claim. Ask: does this robot need to run? Or does it just need to work? Prioritize APIs that expose real-world performance—battery drain, error rates, maintenance cycles—over sprint times. The fastest robot isn’t the most useful. The most useful robot is the one that doesn’t break.
Speed is seductive. It’s visible. It’s quantifiable. But it’s not progress unless it serves a purpose. We’ve built machines that can outrun us. Now we need to build ones that can understand us.
Sources: New Scientist Tech, IEEE Spectrum, PitchBook, CMU Robotics Institute, Toyota Research, Magazino Annual Report 2025
