2 meters. From a standstill.

At 14:41 on August 17, Unitree released a demo video that sent shockwaves through the robotics world: the brand-new humanoid robot "Superhuman" squats and leaps 2 meters straight up. If that number doesn't land for you, the human world record for standing vertical jump—set by Norwegian athlete Sturne Rude in 2025—is 1.83 meters. Unitree's robot jumped 17 centimeters higher than the best a human has ever done.

Even more staggering is the running speed: a top speed of 12.66 m/s, which converts to 45.6 km/h. When Usain Bolt set the 100m world record, his average speed was roughly 10.4 m/s with a peak of around 12.4 m/s. In other words, Unitree's robot already outruns the peak velocity of the fastest human ever.

Leg length: 0.85 meters. Development time for the entire robot: just over three months.

What Those Extra 17 Centimeters Mean

Let's look at the numbers. The "Superhuman" robot's two core metrics both break human world records: 2m standing vertical (human record 1.83m) and 12.66 m/s top speed (human 100m peak ~12.4 m/s).

This isn't "approaching human" or "reaching human level"—this is formally surpassing human biological limits.

This is a landmark moment in embodied AI. For years, humanoid robots we saw either plodded along in labs, performed in carefully choreographed demos, or fell over and couldn't get up. Athletic capability was always humanoids' biggest weakness—balance control, dynamic response, energy efficiency, each a notoriously hard problem.

What Unitree demonstrated shows they've made a critical breakthrough in legged robot motion control algorithms. A 2m vertical jump demands explosive power output and precise landing balance; 12.66 m/s running requires high-frequency gait adjustment and real-time posture feedback. Nailing both simultaneously means their motion control stack has reached a new level.

Even more intimidating is the development timeline: three months. What does that mean? It means Unitree has highly modularized and platformatized the humanoid R&D process—they aren't building a new robot from scratch, but rapidly combining, iterating, and optimizing on top of an existing tech stack. If building a new robot used to be measured in "years," Unitree has compressed the cycle to "quarters."

That iteration speed is the truly chilling part. It can jump 2 meters today—what about in three months? Six months? Next year?

Embodied AI's Motion Singularity Approaches

Unitree's "Superhuman" isn't an isolated event. It's a microcosm of embodied AI's full-on explosion in 2026.

We've already seen a flood of humanoid breakthroughs this year: Boston Dynamics' Atlas doing backflips, Tesla Optimus working in factories, Figure's robot learning to serve drinks, and domestic players like Agibot, Fourier, and Unitree iterating rapidly. But every previous demo still fell within the "approaching human" envelope—movements slightly stiff, reactions slightly slow, prone to falling on tricky terrain.

Unitree just pulled athletic capability straight into the "beyond human" tier. The significance of this leap is on the same level as AlphaGo defeating Lee Sedol—when machines formally surpass humans in a core capability, public perception shifts fundamentally.

Of course, what "Superhuman" demonstrated is pure athleticism so far. Can it jump 2m while recognizing environments, avoiding obstacles, and completing tasks? How's its battery life? Payload capacity? Adaptability to complex terrain? Those questions need more information. Unitree itself notes "there's still significant room for improvement in the coming months."

But this is already enough. Athleticism is the most foundational and critical capability for humanoids—if you can't even walk steadily, run fast, or jump high, all other intelligence is a castle in the sky. Now that the athletic barrier has been crossed, the next step is improving intelligence and generalization.

When robots' physical capabilities surpass humans, many jobs we once thought "robots could never replace" may need reassessment. Delivery, construction, logistics, security, rescue... fields with high physical demands—the timeline for robot replacement may be faster than anyone expected.

See you tomorrow.