On October 13, 2026, China released "Qiusuo" 4.0, the country's first national benchmarking system covering the entire AI industry chain — spanning computing power, model algorithms, intelligent systems and products, industry applications, and safety governance. Led by the China Electronics Standardization Institute (CESI) and related organizations, the framework is designed to give chip makers, model labs, and intelligent-system vendors a unified set of test procedures and evaluation metrics, along with real-world workloads in sectors such as electric power, healthcare, and steel, to solve the industry pain point of incomparable benchmark results across different environments.
The Qiusuo 4.0 system is organized into four sections — foundational hardware and software, models and intelligent systems, safety governance, and industry deployment — and comprises 16 evaluation tools. It covers the full stack from underlying compute, hardware-software co-design, intelligent computing clouds, and AI software engineering, to large language models, agents, embodied intelligence, intelligent terminals, AI safety, and industry-specific scenarios. Bao Wei, director of the model and application innovation research lab at CESI's information technology center, said that lightweight code tools for Qiusuo compute, large-model, and agent benchmarks have been open-sourced, enabling the industry to quickly reproduce evaluations and verify hardware-software coordination, turning abstract benchmark scores into practical selection criteria.
Qiusuo has already completed several rounds of pilot evaluations across the industry chain and product validations in finance, electric power, and oil and petrochemicals, accumulating practical case studies. The release comes as China's AI industry shifts from a "model arms race" toward scenario-driven, engineering-focused deployment. The benchmark is expected to serve as a yardstick in government procurement, industry selection, bidding, and localization substitution, while complementing international benchmarks such as MMLU, HumanEval, and SWE-bench. For enterprises, unified evaluation standards mean lower selection costs and more predictable deployment outcomes; for regulators, they provide a basis for assessing the quality, safety, and compliance of AI products and services.
Top comments (0)