Taobao Flash Sale releases open-source AI large model "Bai Ze": capable of monitoring and analyzing kitchen live streams 24 hours a day, automatically detecting "fake dine-in" situations

IT House, February 28 — Taobao Flash Sale today announced the open-source vertical large model “Bai Ze” (Ostrakon-VL), designed specifically for risk control and governance in the food service and retail store sectors, and announced that the related technical capabilities are now freely available to the entire industry.

According to reports, with the help of the “Bai Ze” large model, industry participants such as food delivery platforms, restaurants, and retail companies can quickly enhance their ability to identify and manage non-compliant scenarios through artificial intelligence technology in areas like image recognition and kitchen alerts, significantly improving digital governance efficiency.

As reported by China Internet Technology, in smart kitchen inspection, “Bai Ze” can analyze live kitchen video streams 24/7, automatically recognizing details such as wall and countertop cleanliness, whether trash bins are overflowing, separation of raw and cooked foods, and whether chefs are wearing hats. After identifying violations, the system can generate alerts instantly and push them to the risk control team, transforming from manual spot checks to continuous monitoring.

To address fake stores, “Bai Ze” uses multi-image comparison and reasoning technology to compare merchant environments with business registration information. The model can automatically flag inconsistencies, effectively curbing violations like multiple store openings under one license and mismatched documentation.

Currently, Taobao Flash Sale supports merchants uploading “one-shot” videos to mark their stores as “dine-in available.” “Bai Ze” can analyze the video content and store information to comprehensively determine whether the store has a genuine dine-in environment. It uses cross-frame consistency analysis to verify video authenticity, preventing deception with static images or AI-generated videos, ensuring consumers receive authentic service experiences.

The model is based on Alibaba Cloud’s Tongyi Qianwen Qwen3-VL-8B foundation, fine-tuned for the specific needs of food and retail store scenarios. So far, “Bai Ze” has been integrated into over 100 production scenarios on Taobao Flash Sale, with over 1 billion calls and token consumption exceeding trillions.

Taobao Flash Sale announced that the technical capabilities related to “Bai Ze” are now open to the industry for free. Model weights and technical reports have been published on mainstream open-source platforms under the Apache 2.0 license.

On February 26, the State Administration for Market Regulation issued new regulations for online food services. Taobao Flash Sale responded promptly, actively implementing these regulations and integrating them into various aspects of platform operations to effectively enhance the safety and trustworthiness of food delivery.

Taobao Flash Sale stated that under the guidance of regulatory authorities, the platform will continue to build a “3+1+AI” food safety co-governance system in collaboration with ecosystem partners and society. Additionally, it will introduce the food delivery rider “Sui Shou Pai” (Handy Snap) as a supplementary force for mobile governance.