X Square Beats Figure AI's Benchmark by 45%: China's Humanoid Compression Hits the Leaderboard Again

X Square specifically targeted Figure AI's published benchmark on robotic manipulation tasks - then claimed to have beaten it by 45%. Combined with Wall-B sorting at 70% lower hardware cost, LimX Dynamics matching Figure's capabilities, and Tiangong Omni earning full marks on work tasks, the pattern is structurally clear: Chinese humanoid labs are running a capability-per-dollar compression that mirrors what DeepSeek did to GPT-4.

In plain terms

Figure AI is one of Silicon Valley's best-funded robotics companies. X Square is a Chinese competitor that specifically used Figure's own published benchmark to test their robot - and claims to have beaten it by 45%. This is the third or fourth data point in a pattern: Chinese humanoid labs are systematically outperforming Western counterparts on cost and, increasingly, on published capability metrics.

The benchmark challenge

X Square's approach - explicitly targeting a competitor's published benchmark - is a specific competitive signal. It says: we believe we can beat their number and we want the market to know it.

Pandaily's reporting makes the claim verifiable with specific numbers: X Square sorted 1,816 parcels in one unedited hour using simple grippers. The target it set for itself - 1,248 - happens to be Figure AI's sustained hourly average, to the parcel. That's 1,816 / 1,248 = 45.6% above Figure's benchmark, in an unedited single-take video.

Two things make this claim stronger than typical benchmark assertions: the task (parcel sorting) is concrete and measurable, and the "unedited" framing is a direct response to the editing criticism often leveled at humanoid robot demos.

The week in humanoid context

X Square's claim didn't arrive in isolation. The same period brought: Tiangong Omni earning full marks on standard work tasks (1.35m, 39kg, full-task benchmark); LimX Dynamics demonstrating a robot rivaling Figure's capabilities at substantially lower hardware cost; WRC 2026 showing 300+ active companies in China's embodied intelligence ecosystem.

The competition loop is running faster than the announcement cycle.

What this means for Figure

Figure's response to this kind of benchmark pressure is consequential. Options: release new results demonstrating performance the challengers haven't matched; shift the competitive frame from benchmarks to deployment volume (harder to fake, more commercially meaningful); or accelerate the roadmap on specific capabilities where the gap is real.

The worst outcome for Figure: being pulled into a benchmark race it didn't design, on metrics that favor the challengers, while deployment customers care about different things entirely.

So what

The humanoid capability gap between China and the US is closing faster on benchmarks than on deployed fleet size. The metric that matters more for market position is real operational hours in real environments - not parcel counts in a demo. Figure's deployment count is the number to watch in the next six months, not its benchmark score.