AINEWS
Three robots from three different manufacturers share one AI base model inside a Guoda Pharmacy (a major state owned Chinese drugstore chain) store, no renovation required.
At a Guoda Pharmacy in Shanghai, three robots built by three different manufacturers are now working the same overnight sort, picking, and delivery shifts, all running on a single AI base model. The store added no new shelving, widened no aisles, and called in no contractors. That last detail is the news: embodied AI, the field that teaches software to drive physical robots rather than only process text or images, has landed in a real retail site without a renovation crew.
The deployment is the public-facing debut for Ant Lingbo, the embodied-AI arm of Ant Group, and a curated exhibit that just won a 镇馆之宝 ("Treasures of the Exhibition Hall") honor at the World Artificial Intelligence Conference (WAIC) 2026, which opens in Shanghai on July 17. The annual committee award, drawn from hundreds of applicants and capped at ten winners, weighs technical content, market prospects, replicability, and social value. It is industry recognition, not a market-share claim. The smart-pharmacy entry was built with state-owned drugstore chain 国大药房 (Guoda Pharmacy) and pairs Ant Lingbo's model with two hardware partners, Leju (乐聚) and Starmap (星海图), alongside Ant Lingbo's own R-2 unit.
Ant Lingbo's LingBot-VLA 2.0 is positioned as a "general-purpose" base model, pretrained on roughly 60,000 hours of real-world robot data and designed to drive 17 manufacturers' hardware across more than 20 configurations: single-arm, dual-arm, bipedal, and wheeled. The pitch is the same one the embodied-AI industry has chased since 2024: stop rebuilding the brain for every new body. If a single base model can transfer across robot brands without scene-specific retraining, the per-site deployment cost, which is the field's main scaling bottleneck, drops with it.
The 17-vendor and 20-configuration figures come from Ant Lingbo via Lei Feng Wang. Open-source coverage of the model release ran in 36kr and 163, which independently confirm that the weights went public but not the operational numbers. Pickup rate, sort accuracy, error rate, store-hours coverage, and pharmacist time saved are not in the public record. The WAIC booth lineup of Leju, Starmap, and R-2 is a curated show-floor display, not a randomized deployment, and the Shanghai Guoda store is the only named operational site.
Two parts of the claim do hold up to a second reading. First, the no-renovation detail is concrete and verifiable: existing shelving, existing lighting, existing floor plan. That is a real change in how retail robotics has historically been rolled out. Second, the partnership between Ant Lingbo and Guoda Pharmacy — pairing a base model provider with a retail operator — represents a division of labor that is closer to how mature software supply chains look than how pilots usually do. Whether the abstraction holds across more than one store is the next test, and Ant Lingbo has not yet published a second site.
The watch item is dated: WAIC opens July 17, and Ant Lingbo has framed its 镇馆之宝 slot as a launch event. The release of the LingBot-VLA 2.0 weights to the open-source community, independently corroborated by multiple trade outlets, is the part of the claim that survives skepticism. The deployment-economics claim, that one base model can drive a real store on multiple robot brands with no construction crew, is the part the next quarter will test.