Spirit AI beats Nvidia on RoboArena robotics benchmark



TL;DR

Chinese startup Spirit AI tops the RoboArena leaderboard Nvidia co-built, scoring 1,924 to Nvidia’s 1,881 as physical AI becomes the next tech battleground.

Two days. That is how long Nvidia’s latest robotics model sat at the top of the RoboArena leaderboard before a startup from Hangzhou knocked it off.

On Wednesday, Spirit AI announced that its foundation model for embodied intelligence, Spirit v1.6, had scored 1,924 on the benchmark, edging out Nvidia’s Cosmos3-Nano-Policy at 1,881. A second Nvidia project, DreamZero, came third with 1,763. It is the first time a Chinese model has claimed the top spot on RoboArena, a benchmark Nvidia co-developed with Stanford University and the University of California, Berkeley.

The timing stings. Nvidia had launched its Cosmos 3 omnimodel at Computex in Taipei on 1 June, calling it the “open frontier foundation model for physical AI.” Trained on 20 trillion tokens of multimodal data, Cosmos 3 was supposed to demonstrate the company’s dominance in a category it practically invented. Spirit AI had other plans.

What physical AI actually measures

RoboArena does not test chatbot fluency or image generation quality. It evaluates how effectively a generalist robot policy translates into real-world actions: object manipulation, navigation, tool usage, perception, planning, and adaptability in unfamiliar environments. In short, it measures whether a machine can think and then do.

Physical AI relies on two core capabilities. Policy capabilities determine a model’s ability to act on what it observes, which is exactly what RoboArena measures. World capabilities determine a model’s ability to simulate and predict what happens next if a specific action is taken.

The industry is moving towards integrating both. Last September, Chinese researchers introduced a unified “Policy World Model” architecture that merges world modelling and trajectory planning into a single system. That convergence is now accelerating across the sector.

China is leading on multiple fronts

Spirit AI’s RoboArena result is not an isolated data point. Across the broader ecosystem of physical AI benchmarks, Chinese firms hold leading positions in nearly every category.

On the WorldArena benchmark, which evaluates embodied world models, the top spot belongs to WorldScape-0.2 from Manifold AI. It outperforms Nvidia’s Cosmos-Predict 2.5 in the policy evaluator track. The perception track is led by AgiBot, one of China’s largest robotics firms, with its GenieEnvisioner-Sim2.0-2B model unveiled last week. The data engine track is topped by DexForce, another Chinese startup.

On the WorldScore benchmark, which tests a model’s ability to generate worlds from text prompts, Manifold AI’s WorldScape-0.2 again leads, outperforming WonderJourney, a joint project from Stanford and Google.

The money pouring in

The technical results are underpinned by an extraordinary wave of capital. Spirit AI announced a 1.5 billion yuan ($222 million) financing round on Wednesday, its fourth in just three months. That pace is reportedly the most aggressive fundraising seen in the embodied AI sector. Earlier rounds had already pushed the company’s valuation past 10 billion yuan ($1.4 billion).

On the same day, XYZ Embodied AI, incubated by the Beijing Academy of Artificial Intelligence, said it had closed its pre-A round, having raised 1 billion yuan in just 10 months to develop “embodied brains” and world models. Manifold AI has completed five funding rounds in 10 months, with its latest in April securing reportedly hundreds of millions of yuan.

The broader Chinese robotics sector attracted $3.4 billion in venture funding in 2025 alone, 42 per cent more than the United States. That gap appears to be widening in 2026.

Nvidia’s counter-strategy

Nvidia is not standing still. At Computex, CEO Jensen Huang announced a partnership with Chinese robotics firm Unitree, which is preparing a $7 billion IPO, and Singaporean robotic hand maker Sharpa to build a humanoid robot reference design. The platform combines Unitree’s H2 Plus humanoid body, Sharpa’s Wave tactile hands, and Nvidia’s Jetson AGX Thor T5000 processor.

Huang also launched the Cosmos Coalition, enlisting AI labs including Agile Robots, Black Forest Labs, Runway, and Skild AI to advance open world models. The message is clear: Nvidia wants to be the infrastructure layer for the entire physical AI ecosystem, even if individual models lose benchmark crowns.

But Huang himself identified the sector’s fundamental bottleneck. “For robotic systems and physical AI, data is the hardest problem,” he said at Computex. That admission points to why China may hold a structural advantage.

The data question

Alexandr Wang, the Scale AI founder who joined Meta as its first chief AI officer in 2025, reportedly said last year that China was “fundamentally very well positioned on data” and that many US companies relied on Chinese data to train robotics foundation models. The claim has not been independently verified in full, but it aligns with observable policy.

In tech hubs like Beijing and Shenzhen, Chinese authorities have established state-backed “data factories” to collect robotics training data at scale. China’s manufacturing supply chain, with its dense network of factories and assembly lines, provides a natural environment for the kind of real-world robotic interaction data that remains expensive and difficult to gather in the West.

If data really is the hardest problem, the country with the most factories, the most robots, and the most willingness to collect may already have the answer.



Source link

Leave a Reply

Subscribe to Our Newsletter

Get our latest articles delivered straight to your inbox. No spam, we promise.

Recent Reviews


Robot mowers on a yard

Maria Diaz/ZDNET

Follow ZDNET: Add us as a preferred source on Google.


The perfect robot mower for you is not nearly as fancy and feature-heavy as you may think. I’ve said it before, and I’ll say it again: it’s not the lawn mower, it’s all about the yard. A robot mower may be a market leader with top-of-the-line specs and still not be a good fit for your yard.

Here’s the great news: There’s a perfect robot mower for almost any yard. As someone who’s tested numerous types of robot lawn mowers, I’ve learned that many of the specs that brands market as groundbreaking are simply not vital for most shoppers. A mostly flat, fenced-in 0.10-acre yard doesn’t need the power that a hilly, sectioned, unfenced one-acre yard does.

Also: I tested the Ferrari of robot mowers for a month – here’s my verdict

If you’re looking to choose the best mower for your home, be sure to check out ZDNET’s robot mower buying guide

Here’s what you don’t need to stress over when buying a robot mower

Eufy E15 Robot Mower

Maria Diaz/ZDNET
For yards with… Best robot mower type Examples
No fences A wired boundary is best, but a great GPS/RTK robot mower can stick to the map you make with it. Yardcare E400, Mammotion Luba 3
Fences A LiDAR robot mower that can be dropped to mow with little setup and learn its map as it navigates. Eufy E15, Ecovacs Goat A3000
A lot of trees A LiDAR or wired boundary mower, since trees can interfere with satellite signals. Husqvarna iQ series (optional wire, EPOS)
Unbordered garden beds A GPS/RTK robot mower that you can set up to avoid flower beds when mapping. Mammotion Luba 3, Husqvarna iQ Series
Bordered garden beds A LiDAR, GPS, or wired boundary robot mower works for these yards. If you choose a wired boundary, you may have to bury wire around the flower beds, unless the borders are tall enough for the mower to avoid. Mammotion Yuka, Navimow Series H
pets A LiDAR robot mower that can adjust its navigation in real-time in reaction to its surroundings. Mova LiDAX Ultra 2000, Segway Navimow i2
Hills and uneven terrain An AWD robot mower capable of handling steep slopes, regardless of the navigation type. Mammotion Luba 3, , Husqvarna iQ

1. Don’t focus on: ‘AI-powered’ or other marketing buzzwords

Segway Navimow X3 Series robot mower

Maria Diaz/ZDNET

Artificial intelligence (AI) has surpassed the popularity of acid-wash jeans in the 80s and Baby G watches in the early 2000s. And tech companies — including robot lawn mower manufacturers — are capitalizing on its appeal.

Most of these “AI-powered” or “intelligent mowing” terms are vague, geared to grab shoppers’ attention with buzzwords. That doesn’t mean that the robots don’t use AI to navigate, however. 

The key is to find out how the robot uses AI to its benefit, and whether that will meet your AI expectations. 

Also: This robot mower took care of my lawn for months – and it’s currently $300 off

AI algorithms typically process data captured by the robot’s hardware to help it make quick decisions and adjustments. For example, a robot lawn mower may have a set of sensors and cameras to capture its surroundings. The robot’s processor then uses AI to convert that information into actionable data, so it knows whether to swerve to avoid an obstacle or slow down around a retaining wall.

Instead, look for: The navigation tech under (and on) the hood

Instead of AI and other buzzwords, you should focus on matching the robot lawn mower’s hardware and navigation system to your yard. This includes whether the robot uses RTK (Real-Time Kinematic) for positioning, and whether it features LiDAR, cameras, and sensors. 

Then look at real user reviews to assess how accurately the robot mower maps and how well it performs around various types of obstacles.

There’s no blanket rule for robot mowers, but most do well with the following guidelines.

2. Don’t focus on: Premium extras

Yardcare E400 robot lawn mower

Maria Diaz/ZDNET

Skip the premium extras that don’t match your yard. You really don’t need the most advanced robot mower; you need the one that will best handle your lawn. 

Most US homeowners have mostly flat lawns, simple rectangular layouts, minimal obstacles, and small yards. Yet some of the most popular mowers advertise features that don’t match this, and you don’t want to spend an extra few hundred dollars on advanced features that won’t deliver a noticeable difference in your yard.

Instead, look for: Only as much as you need

Do you have a mostly flat lawn with no fences and need a robot that can navigate to several sections separated by paths? Then you can skip AWD models and commit to superior mapping and navigation features, like multi-zone intelligence.

Also: I let a modular yard care robot mow my lawn – here’s my verdict after a month

Similarly, if you have a yard with dense trees covering most of it, it’s safe to skip the RTK models and go for LiDAR or boundary wire options instead. 

3. Don’t focus on: Flashy app features

Mammotion Luba 2 robot mower path

The path lines created by the Mammotion Luba 2, as captured by our Bink Outdoor camera, is one flashy app feature I can’t quit.

Maria Diaz/ZDNET

Any dependable robot lawn mower requires an equally reliable mobile app to let you use it effectively. However, manufacturers market many flashy app features that end up being unnecessary for many users. 

Don’t make app features the deciding factor unless it’s something you genuinely care about. Many users don’t rely on voice control to run their mowers and don’t mind using a separate app for their robot rather than integrating it into an existing home automation system.

Also: I let a smart planter maintain itself for 2 months – here’s the result

A robot lawn mower with mediocre navigation and cutting performance can still have a flashy app — all while leaving behind missed patches or taking longer to finish mowing.

Instead, look for: The features you’ll actually use

Most robot mower users keep them running on a schedule to get the lawn-cutting chore off their minds. The majority of the most popular models offer basic features beyond scheduling, such as remote start and stop, basic mapping, automatic rain delay, and theft protection. 

It’s easy to find robot lawn mowers with these features, but if you’re looking for anything beyond that, just be sure that the feature is worth it, especially if you’re paying extra for that model.

Also: I’ve tested robot mowers for years – here’s my expert advice for every yard type

An example of a flashy app feature that is completely unnecessary, but I love having? The Mammotion’s pattern cutting. I can select the cutting pattern I want on the Mammotion app, whether I want lines or checkered, but I can also have the robot cut in custom patterns, like letters and numbers. I don’t care for mowed letters in my yard, but I like that it always has that freshly mowed checkered patterned with no effort from me. 

4. Don’t focus on: Cutting system extras

Segway Navimow X3 Series robot mower

Maria Diaz/ZDNET

The cutting width and system specs are important, as they can determine whether a robot can cover a given area in a day. However, most robot mowers use similar multiple-blade mulching systems. 

Unlike traditional lawn mowers with large blades for aggressive cutting in a single pass, robot mowers typically feature a set of small blades that constantly spin. Because of this, robot mowers trim smaller amounts of grass with each pass than a traditional mower, but they also cut more frequently and leave behind smaller grass clippings that decompose naturally.

Also: I powered my 3,000-sq-ft home with an EcoFlow battery in a blackout – here’s how it kept my AC on

Because the robot mowers have a smaller, compounding cutting system, the real-world differences between the cutting systems from one brand to another are often smaller than you’d expect. Other issues, like poor navigation, will be glaringly obvious before small differences in blade design.

Instead, look for: Cutting width and yard size

The average US yard would benefit more from navigation quality, consistency, and connectivity than blade design. Instead, you should focus on matching the mower to your yard size.

The robot’s capacity is measured in how many acres it can cover in a day. Among other features, this is calculated based on your robot’s battery size and cutting width. Essentially, most users want a robot that can mow an entire yard in a day, so you can set it and forget it and always come home to a mowed yard. You get this by getting the appropriate robot for your yard size.





Source link