Version 4.0 of the "QiuSuo" evaluation benchmark, the first in China targeting the full artificial intelligence industry chain, was officially released on October 13. The framework covers computing power, model algorithms, intelligent systems and products, industry application, and security governance.
QiuSuo is a national standard AI evaluation benchmark system that gives chip and model vendors a unified testing process and yardstick. It is bundled with real business test workloads from sectors including finance, power, and steel, addressing the industry pain point that evaluation results are not comparable across different environments.
The system has already run multiple rounds of pilot evaluations across the industry chain, completing product validation in finance, power, oil and petrochemical scenarios and accumulating a number of industry implementation cases.
Bao Wei, director of the Model and Application Innovation Research Office at the Information Technology Center of the China Electronics Standardization Institute, said lightweight code tools for QiuSuo's computing power, large model and agent modules have been open-sourced so the industry can quickly reproduce evaluations and conduct software-hardware co-verification. The aim is to turn abstract evaluation scores into a real basis for procurement decisions.
The full system is divided into four blocks covering basic software and hardware, models and intelligent systems, security governance, and industry implementation, with 16 measurement tools. Coverage extends from underlying computing power, software-hardware coordination, intelligent computing cloud and intelligent software engineering, to large models, agents, embodied intelligence, intelligent terminals, AI security and industry scenarios.
The release marks a move toward standardizing how AI systems are measured, an area industry observers say has been fragmented, with different vendors reporting results under incomparable conditions.
Source: CCTV News / People's Daily
