Ant Group's LingBot Tops 2.2 Million Downloads in Nine Months
Ant Group says its open source LingBot robot models have been downloaded more than 2.2 million times and collected over 33,000 GitHub stars in the nine months since their first release, and that robots running them are now picking medicines in Shanghai pharmacies in normal trading hours.
The embodied AI unit presented a full stack 2.0 lineup at the World Internet Conference in Wuzhen this week, split across perception with LingBot-Vision and LingBot-Depth, world models with LingBot-Video and LingBot-World, and action models with LingBot-VLA and LingBot-VA. Six models, one stack. The design goal is one model reused across different robot bodies and tasks instead of training a separate brain for each machine.
But the more telling number is who else is using it. Ant says the work has been cited by Physical Intelligence and Generalist AI, that Stanford's OpenWAM project adapted the LingBot-VA module, that a Korean project called FastOPD used LingBot-VLA as a teacher model to train lighter policies, and that EPFL used LingBot-Map to reconstruct coral reefs in 3D.
The tooling has spread the same way, which is how a research lab measures whether a model is actually being used rather than only read. LingBot-VLA and LingBot-VA have been adapted to Hugging Face's LeRobot platform, LingBot-Vision was merged into the main branch of the timm vision library, and LingBot-World runs on both vLLM and SGLang, with NVIDIA integrating it into its open source world model library FlashDreams.
And the competition record backs the adoption claim. The first LingBot embodied model challenge drew 1,855 teams and 3,252 participants from 648 universities and companies across 13 countries in its online preliminary round.
But the commercial deployment is narrower than the download figures suggest. At the summit Ant and Guoda Pharmacy showed a robotic pharmacy that recreates a real store with densely stocked shelves, where a robot works an aisle 80 centimetres wide and handles the flow from receiving an order to picking and handing the item over.
That detail matters more than it looks. Retail aisles are built for people, and a machine that fits an 80 centimetre gap without the store rearranging its displays is a machine that can be installed in existing shops rather than in a showroom built around it. Pharmacies are the hardest version of that test, because the shelves are dense and the products are small.
So the pharmacy is the proof of a specific claim rather than a rollout. Ant says the setup is in regular operation at several Guoda stores in Shanghai, with no changes to product displays or store layout, taking repetitive picking off pharmacists, and the company is separately testing industrial loading and logistics sorting with partners who have not been named.
And beyond China the team has signed a memorandum of understanding with the Arab Federation for Digital Economy to explore retail, logistics and hospitality uses in the Middle East, which is the same open model strategy carried into a market where the alternative would be a locally trained stack built from scratch.
Ant says LingBot models passed 2.2 million downloads and 33,000 GitHub stars in nine months
The 2.0 lineup covers perception, world and action models under one stack
Stanford, EPFL, Physical Intelligence and Generalist AI are named as users or citers
Robots work 80 centimetre aisles in Guoda pharmacies in Shanghai in regular operation
Ant has signed an agreement to explore retail and logistics uses in the Middle East
Source: Pandaily
