The company behind China’s Alipay just gave away an AI that builds a playable video game world and keeps it running for a full hour. Robbyant, the robotics arm of fintech giant Ant Group, open-sourced its LingBot-World 2.0 world model this week. It generates an interactive 3D world in real time, at 720p and 60 frames per second, that a user can walk around and act inside.
That release was not a one-off. It was the finale of a remarkable week. Across four days, Robbyant open-sourced an entire robot brain, piece by piece: the eyes, the hands, and now the imagination. For a company most people know through a payments app, it is a bold land-grab in embodied AI.
An hour-long world you can play
Start with the headline act. Most AI video tools generate a clip and stop. LingBot-World 2.0 keeps going. The company says it holds visual quality for a full hour with no drift, where rival systems usually blur and collapse after a while.
It is also interactive, not just watchable. On Robbyant’s Reactor platform, a user steers a character with the keyboard as the scene renders live. The action list reads like a game: attack, shoot arrows, cast spells, jump, and glide. Type a command and you can flip the world to night, change the weather, or drop in a new object.
Two built-in agents run the show. A “pilot” plans and carries out the character’s moves. A “director” throws in new events as the scene unfolds. The world even supports several people at once in one shared, persistent space. Robbyant also open-sourced a second model, LingBot-Video, which it calls the first open-source video generator built for robots rather than film.
The 💜 of EU tech
The latest rumblings from the EU tech scene, a story from our wise ol’ founder Boris, and some questionable AI art. It’s free, every week, in your inbox. Sign up now!
A whole robot brain, given away
The world model is the flashy part. The rest of the week was arguably more important for the robotics industry.
On one day, Robbyant released LingBot-VLA 2.0, a “vision-language-action” model it pitches as a universal brain for robots. The company trained it on 60,000 hours of real-world data, drawn from 20 kinds of robot across 17 makers, including Unitree, Fourier, and AgiBot.
It claims the model beats Nvidia’s GR00T and Physical Intelligence’s π0.5 on a shared academic benchmark. It also runs with under 130 milliseconds of latency on a single consumer graphics card.
A day earlier came LingBot-Depth 2.0, aimed at a problem that has long haunted robots: seeing glass, mirrors, and shiny surfaces. Depth cameras tend to fail on them, which is how a robot walks into a glass door.
Robbyant says its model halves the error of its predecessor in the hardest indoor scenes. A paired vision model, it adds, matches strong systems while training on a fraction of the images that Meta’s DINOv3 used.
Why give it all away
The obvious question is why a fintech giant hands this out for free. The answer is strategy. China has more than 100 humanoid robot makers, and most build the body, not the brain. Whoever supplies that brain becomes the layer everyone else builds on.
Open-sourcing is how you get there fast. Give away the models, win the developers, and set the standard before a rival does. It is a pointed contrast with Nvidia, whose GR00T platform is open but tied to its chips.
It also fits a wider Chinese pattern, from MiniMax to Alibaba, of flooding the field with free, capable models.
There is a national angle too. China is racing to lead embodied AI, and it has already started to register its humanoids by ID. A shared, home-grown software stack helps its whole robot sector move together.
From Alipay to household robots
Robbyant sits inside Ant Group, the Hangzhou company that runs Alipay and grew out of the Alibaba empire. Ant is not a robotics firm by history. It is a payments and finance business, and a huge one.
It is spending like it wants to change that.
Ant reported record research spending of $5.17bn last year. Robbyant’s stated goal is not factory arms but robotic companions and carers, for elderly care, medical help, and chores at home. A robot that minds an ageing parent has to see a glass table, grip a cup, and understand a messy room.
This week’s releases are the parts for exactly that.
Why it matters
Robots have long been held back less by motors than by minds. They could move, but they could not reliably understand a room or plan their way through it. Robbyant is betting that better perception, a shared brain, and world models to practise in will crack that.
The claims are the company’s own, and self-reported benchmarks deserve caution until others test them. But the direction is clear enough. One of China’s most powerful tech groups just put a near-complete embodied-AI toolkit into the open, in a single week, for anyone to use.
The race to give robots common sense now has a very well-funded new entrant.
