- inclusionAI posted training-content summaries for 11 model versions across three families on Hugging Face, covering Ling-2.0 through Ling-3.0-VL, Ring-2.0 through Ring-2.6, and Ming-Omni plus two 16B-A3B multimodal variants.
- The documents follow the public training-content summary template linked to Article 53(1)(d) of the EU AI Act, signaling preparation for overseas compliance rather than domestic regulatory filing.
- The repository explicitly states it contains documentation only — no model weights, training datasets or third-party content rights — and that publication does not constitute certification or authority endorsement.
What Was Published
On September 24, Ant Group's inclusionAI opened an AI-Transparency repository on Hugging Face containing structured training-content summaries for its major model lines. The documents cover Ling foundation models from version 2.0 through 3.0 (including LLM, flash, tiny, flash-fin and vision-language variants), Ring reasoning models from 2.0 through 2.6-1T, and the Ming family including Ming-Omni, Ming-UniAudio-16B-A3B and Ming-UniVision-16B-A3B.
Each summary is scoped to the specific model names and versions identified in the document. The repository notes that a model-family name, quantization suffix or collection membership does not extend a document's scope beyond what it explicitly names.
Format and Limitations
The disclosures adopt the standardized format tied to Article 53(1)(d) of Regulation (EU) 2024/1689, which requires general-purpose AI model providers to publish information about their training content. This makes the release a structured compliance-oriented filing rather than an informal technical blog post.
The repository states plainly that it holds documentation alone, without model weights or training datasets, and that the documents do not distribute underlying training data or grant rights to third-party content. It also clarifies that publishing these files does not amount to certification, authority endorsement, or confirmation that every inclusionAI model is covered.
Context for Chinese AI Labs
As global regulators move to make training transparency a mandatory requirement rather than a voluntary practice, Chinese frontier labs face growing pressure to document their data provenance for overseas markets. inclusionAI's decision to publish structured summaries aligned with the EU framework signals an effort to prepare for international compliance obligations while keeping proprietary training data and model weights separate from the disclosure.
Sources & context
Go to the original material. Company claims remain attributed to their sources.
01
