Tiangong: How China Built a State-Backed, Open-Source Humanoid

In the West, a humanoid robot is a company's crown jewel — Tesla's Optimus, Figure's 03, guarded behind patents and closed source. China built a different kind of thing entirely. Its highest-profile humanoid, Tiangong (天工), isn't a product a startup sells; it's a platform a state-backed innovation center gives away — the running gait, the datasets, even the AI models — while the same center writes the national standards everyone else will have to follow.

That center is the Beijing Humanoid Robot Innovation Center, branded internationally as X-Humanoid. It's the outfit whose robot won the world's first humanoid half-marathon, and it operates on a logic that looks strange from Silicon Valley: seed the entire industry with free, shared technology, and set the rules of the game in the process. Here's how that open-source, state-backed playbook actually works — and why it might matter more than any single flashy robot.

Not a startup — a national platform

The first thing to understand is that X-Humanoid isn't trying to be Tesla. Founded in November 2023 in Beijing's E-Town (Yizhuang) — China's densest robotics cluster — it was set up as a shared "mother platform," and in October 2024 China's Ministry of Industry and Information Technology and the Beijing government jointly designated it a "national-local co-built" embodied-AI robot innovation center. Translation: it has an explicit state mandate to build common technology the whole domestic industry can draw on.

Its shareholders read like a who's-who of Chinese robotics. UBTech, Xiaomi Robotics and the Beijing state enterprise Jingcheng Electromechanical each hold about 28.57%, alongside E-Town's Yizhuang Robotics vehicle, and UBTech is the lead unit — its CTO, Xiong Youjun, is also the center's CEO. In February 2026 the center raised more than 700 million yuan in its first market round, with Baidu among the strategic investors. It's the institutional opposite of a garage startup: a consortium of giants plus the state, pooling effort so the whole country's robot industry moves at once. That's the same state-coordinated instinct we traced in China's push to shape global robot regulation and its nationwide robot ID scheme.

"Silicon Valley builds a robot to own it. Beijing built a robot to hand it out — and to make sure the whole industry standardizes on its version."

Tiangong the robot: from a 6 km/h jog to a Thomas flare

Tiangong itself has moved fast. The original Tiangong (April 2024) — 163 cm, 43 kg — was billed as the world's first pure-electric full-size humanoid able to run "anthropomorphically," holding a steady 6 km/h. Tiangong 2.0 (early 2025) pushed that to 12 km/h, and the marathon variant, Tiangong Ultra (1.8 m, ~55 kg), turned running into a headline (more on that below).

The current flagship, Embodied Tiangong 3.0, was unveiled in February 2026: 1.69 m, 62 kg, 43 degrees of freedom, with peak joint torque of 400 N·m. In its debut it did somersaults, a gymnastic Thomas flare, parkour and a one-handed vault over a roughly one-metre obstacle — the center calls it the first full-size humanoid with "contact-interactive whole-body high-dynamic motion control." Crucially, it's meant to leave the demo stage: scaled mass delivery is targeted for the second half of 2026, built around D-Robotics' Sunrise S600 compute chip, with the center projecting a more than 50% cut in total unit cost. No retail price has been announced. For how that fits the broader race to build these machines at volume, see who's actually mass-producing humanoids.

The race that made it famous

If you've seen exactly one thing about Tiangong, it's probably the running. On April 19, 2025, Beijing's Yizhuang district staged the world's first humanoid-robot half-marathon — 20 robot teams on a 21.1 km course alongside some 12,000 human runners. Tiangong Ultra won it in 2 hours 40 minutes 42 seconds, one of only six robots to finish, swapping its battery once along the way. Second place went to the N2 from Noetix, the startup behind the $1,400 Bumi, the cheapest walking humanoid.

One honest caveat, because evergreen accuracy matters: a second edition of the race in 2026 was won by a different, far faster robot in around 50 minutes, so Tiangong's win was of the inaugural event, not a standing record. But the point of the first race wasn't speed — it was proof that a full-size electric humanoid could run 21 km at all. That's exactly the kind of real-world stress test we argue for in how to actually test a humanoid robot: a backflip is a demo, a half-marathon is a benchmark.

The real strategy: give the whole stack away

Here's where X-Humanoid diverges hardest from the Western model. Instead of hoarding its technology, it runs a Tiangong Open Source Plan and publishes the pieces other companies need most:

The strategy is shrewd. Training data and reliable locomotion are the two things every humanoid startup lacks; by giving them away, X-Humanoid makes its stack the default that a hundred smaller Chinese firms build on — which quietly turns its choices into the industry's de facto standard. It's the same "own the platform, not the product" logic behind the software layer we covered in Leju's Kuavo and its Huawei-born OS, and it maps onto the different AI philosophies powering the humanoid race.

And then it writes the rules

Open-sourcing the technology is only half of it. The center is also a rule-writer. It led China's — and it claims the world's — first "Humanoid Robot Intelligence Grading" standard (T/CIE 298-2025), a framework that sorts humanoid capability into five levels from L1 to L5, much like the autonomy levels used for self-driving cars. And it is a lead drafter of China's first batch of national humanoid-robot standards, taking charge of the motion-control section (with UBTech leading the task-operation part), plus general safety requirements.

Put the pieces together and the design of the whole thing becomes clear. A body other firms can copy, a brain and datasets they can download, and the standards they'll be measured against — all coming from one state-backed hub. Whoever sets the shared base and the yardstick shapes the industry, and in humanoids that's increasingly China, a theme running through why robots are the next global trend after AI.

"Tiangong is a robot, but X-Humanoid is really building something bigger: the shared foundation — code, data, and rules — that an entire national industry stands on."

Why it matters

It's tempting to file Tiangong under "another impressive Chinese demo," but that misses the point. The robot is the marketing; the platform is the strategy. By open-sourcing the hard parts and authoring the standards, China is trying to do for humanoids what Android did for phones — become the base layer everyone builds on. Western humanoids may well be more capable today, but they're islands. Tiangong is infrastructure. When you're weighing how soon humanoids actually arrive, that difference — a shared, state-funded foundation versus a dozen closed silos — may end up mattering as much as any single robot's backflip.

Frequently asked questions