Ant Financial's inclusionAI has opened a repository on HuggingFace called AI-Transparency, publicly releasing summaries of the training content for several major model series, including Ling, Ring, and Ming. This is not another round of releasing model weights, but rather an active disclosure regarding the question of "what data did the model actually train on." The repository contains only documents, no weights, and no training datasets, but clearly outlines the origins of models from Ling-2.0 to Ling-3.0, Ring-2.0 to Ring-2.6, as well as Ming-Omni and two 16B-A3B multimodal models.

The framework of this summary itself carries regulatory implications: it strictly follows the "public training content summary template" associated with Article 53(1)(d) of Regulation (EU) 2024/1689 of the EU Artificial Intelligence Act. In other words, this is not a self-written technical blog post, but rather a standardized format for disclosing training content in accordance with the compliance requirements imposed by the EU on providers of general-purpose AI models. The scope of each document is strictly limited to the specific model and version it mentions - if a document mentions Ling-2.0, it only covers Ling-2.0 (including Ling1T, Ling-flash2.0, and Ling-mini-2.0), and does not automatically expand to other models in the same series or those with quantization suffixes.

When the list is laid out, the coverage is quite extensive: along the Ling line, from 2.0, 2.5, 2.6 up to 3.0, where 3.0 is divided into LLM (including flash, tiny, flash-fin) and VL (Visual Language) version (Ling-3.0-flash-VL); Ring from 2.0, 2.5-1T to 2.6-1T; the Ming family includes Omni multimodal (including flash-omni-preview and flash-omni-2.0) as well as UniAudio-16B-A3B and UniVision-16B-A3B, two unified audio and visual models. Each summary is marked with its own version number and update date, describing the training content at the specified level within the corresponding document, without distributing the underlying training data or granting third parties any rights to the content. The official also specifically emphasizes that the existence of these documents does not equate to authoritative certification or endorsement, nor does it mean that every model under inclusionAI is covered.

The contact page points to Ant Financial's official developer site for Ling, while the version history reminds readers that the dates in the PDF are only document version dates and do not represent the initial online release time; actual changes should be referred to the commit records in the repository. As global regulation turns "training transparency" from an optional feature into a hard threshold, Ant's step of systematically organizing and publicly releasing training summaries for the entire model series is both paving the way for overseas compliance and inadvertently opening a small window to show how their models were developed.