en.Wedoany.com Reported - As large model technology accelerates its iteration, tokens, as the smallest information units for input and output in large models, are evolving from behind-the-scenes technical terms into key concepts in the era of the intelligent economy. According to statistics from the National Data Administration, as of the end of March 2026, China's daily token call volume has exceeded 140 trillion, an increase of over a thousandfold compared to early 2024. Tokens are increasingly becoming a core metric for measuring model service capability, computing power utilization efficiency, and industry maturity. High-quality token services bridge model training and inference optimization downstream while empowering application scenarios across thousands of industries upstream, playing a vital role in consolidating the foundation of large model supply and unleashing the value of artificial intelligence applications.

To promote the high-quality development of token services in China, the China Academy of Information and Communications Technology (CAICT), in collaboration with 10 organizations—Huawei Cloud, Ant Digital Technologies, Mobile Jiutian, Unicom Digital Intelligence, JD Cloud, Lenovo Baiying, Inspur Haiyue, Haokj Technology, Tianyun Data, and Paiou Cloud—has jointly launched the "Token Service Capability Climbing Plan" (referred to as the "Climbing Plan") and formulated and released the "Enterprise-Level General Scenario Token Service Performance Climbing Baseline."

First Batch of Enterprise-Level Token Service Performance Climbing Baseline (General Scenario)

Tokens Per Second (TPS) ≥ 55 tokens/second

Time to First Token (TTFT) ≤ 0.9 seconds

Call Success Rate ≥ 99.9%

The "Climbing Plan" focuses on establishing baselines, improving performance, stabilizing services, reducing costs, and building ecosystems as core directions. It targets key difficulties and common issues in the engineering implementation of token services, leveraging monitoring data from CAICT's "Large Model Token Service Performance Monitoring Platform" to provide quarterly dynamically updated performance baselines. It regularly publishes enterprises that meet the "Climbing Plan" standards, driving token services from "usable" to "user-friendly and affordable" stages.

CAICT's Artificial Intelligence Research Institute has long been dedicated to policy research, technology development, standard setting, ecosystem building, and testing and evaluation in the field of artificial intelligence. Moving forward, it will continue to promote the "Token Service Capability Climbing Plan" to accelerate the high-quality development of token services.

First, it will continuously solicit climbing enterprises, welcoming more companies to join and challenge the "Climbing Plan" to collectively enhance token service capabilities in general scenarios. At the same time, CAICT will optimize the tools and methods for climbing tests of token service capabilities to ensure the effective operation of the dynamic update mechanism for climbing baselines.

Second, it will jointly build industry-level climbing plans, collaborating with leading enterprises and technology providers in various industries to construct climbing baselines for token service capabilities tailored to vertical industry scenarios and develop climbing pathways.

Third, it will provide full-stack capability assessments for high-quality token services. On one hand, it will continuously iterate and improve the performance testing platform (TokenBenchX) and dedicated test datasets to offer professional comprehensive performance evaluations for token service enterprises. On the other hand, it will provide full-stack capability assessments for token services, covering the entire process of token service production and operation, as well as service quality and security.

This article is compiled by Wedoany. All AI citations must indicate the source as "Wedoany". If there is any infringement or other issues, please notify us promptly, and we will modify or delete it accordingly. Email: news@wedoany.com