Steel large models received a complete set of national standardisation guidance documents on October 19, 2026. China's State Administration for Market Regulation (the Standardization Administration) recently issued three documents, the AI Steel Industry Application Scenario Classification Guide, the Technical Requirements for Steel Large Model Datasets, and the Steel Large Model Service Capability Maturity Assessment, reported by Xinhua, CCTV News and China Securities Journal on October 19. According to China Securities Journal, more than 50 organisations from research, industry and academia, including steel research institutes, Baowu and Baosight, jointly drafted them.
Each document covers one stretch of the road
The first covers scenarios: for long-process steel enterprises, it classifies and describes AI application scenarios across R&D and design, process control, operations, business management and the industrial ecosystem, helping companies decide which links are worth a large model. The second covers data: it establishes the structure of steel large-model datasets and sets technical requirements for raw data collection, cleaning and governance, knowledge fusion, sample formation and data security. The third covers acceptance: it provides a service capability framework and assessment indicators, with maturity grading rules and evaluation methods. Together they form a complete chain of scenario classification, dataset specification and maturity assessment.
The real difficulty is data and acceptance, not modelling
The steel industry does not lack model suppliers; it lacks reusable data specifications and a comparable acceptance ruler. Data definitions and quality vary widely between plants for the same blast furnace or the same process, so a model that works in a pilot may fail at another plant. The dataset document turns collection, cleaning, fusion and sample formation into unified requirements, effectively paving a road with consistent data specifications for industry models. The maturity assessment gives buyers and regulators a shared language for acceptance, instead of judging by demonstrations alone.
For steelmakers and vendors, this is a shared exam
These are standardisation guidance documents, directional and referential in nature, not mandatory national standards. In practice, however, such documents often become the default reference in bidding and project acceptance: steel companies can map projects to the scenario classification, while vendors must deliver data to the dataset requirements and pass evaluation under the maturity framework. The direct effect is to move AI in steel from isolated pilot stories toward a stage that can be compared horizontally and replicated at scale. What to watch next is which enterprises produce the first steel large models assessed under this framework.
