On December 26, 2025, China officially put into effect its inaugural national standard series, specifically tailored for general-purpose large models. This series, named 'Artificial Intelligence Large Models,' heralds a fresh era of standardized advancement within the large model sector.

Prior to this, there was a notable void in the technical evaluation framework for large models. This newly introduced standard series bridges that gap by clearly defining the prerequisites for performance, safety, and service capabilities. Furthermore, the evaluation tools and methods stipulated by the standards have gained recognition from the China National Accreditation Service for Conformity Assessment (CNAS), ensuring their credibility and reliability.

As of now, these standardized tools have successfully executed over a thousand evaluation tasks, invoking large models more than 950,000 times in the process. Through rigorous testing, they have effectively pinpointed prevalent issues such as hallucination control (where AI generates inaccurate or nonsensical information) and content safety. Additionally, these standards have aided nearly 30 vendors in refining their technologies, fostering a seamless closed loop of 'research and development—evaluation—application—upgrading.' This iterative process not only enhances the quality and safety of AI large models but also propels the entire industry towards a more mature and standardized future.