China on Tuesday released Qiusuo 4.0, the country's first evaluation benchmark covering the entire artificial intelligence (AI) industry chain including computing power, model algorithms, intelligent systems and products, industry applications and security governance.
This national AI standard evaluation benchmark system provides manufacturers of chips, models and other products with a unified testing process and evaluation criteria.
It also includes real-world business test workloads for sectors such as power, healthcare and steel, thereby addressing the industry's long-standing pain point of incomparable evaluation results across different environments.
So far, Qiusuo has conducted multiple rounds of pilot evaluations across the industry chain, completed product validation in scenarios such as finance, power and petrochemicals, and accumulated numerous practical implementation cases, serving as an excellent model for the industry.
"The lightweight versions of Qiusuo code tools including those for computing power, large models, and intelligent agents, have been open-sourced to facilitate rapid replication of evaluations and hardware-software co-verification within the industry. This transforms abstract evaluation scores into concrete criteria for product selection, thereby accelerating the efficient implementation of AI technologies," said Bao Wei, director of the Model and Application Innovation Research Office at the Information Technology Center of China Electronics Standardization Institute.
The entire system is divided into four major sections: foundational software and hardware, models and intelligent systems, security governance and industry implementation.
It comprises 16 evaluation tools, covering comprehensive assessments ranging from underlying computing power, software-hardware coordination, intelligent computing clouds, and intelligent software engineering to large models, intelligent agents, embodied intelligence, intelligent terminals, AI security and industry scenarios.
China releases first AI evaluation benchmark Qiusuo 4.0
