Recently, token economy has become a hot topic of concern in the AI industry. As the smallest unit for processing information in large models, the exponential explosion of lexical tone usage has become the driving force for breakthroughs and large-scale implementation of terminal intelligent agent technology, promoting AI to move from the cloud to terminals and evolve from auxiliary tools to autonomous agents.

According to data, as of March, the daily average usage of keywords and tones in China has exceeded 14 trillion yuan, achieving a growth of over a thousand times compared to 100 billion yuan at the beginning of 2024. The relevant person in charge of the National Data Administration stated that the significant increase in the daily average usage of keywords marks the rapid growth stage of China's artificial intelligence development. At the same time, China's high-quality data supply system continues to improve. By the end of 2025, over 100000 high-quality datasets will be built nationwide, with a total volume of 890PB, laying a solid data foundation for the development of the keyword economy and terminal intelligent agents.

Every time an intelligent agent thinks, plans, and executes, it consumes lexical elements. Faced with pain points such as high cost of word element consumption, strong cloud dependence, and privacy and security risks, terminal manufacturers have taken the lead in using end-to-end intelligent agent technology to solve problems and promote the evolution of word element economy towards "efficiency, low cost, and security". Recently, Honor released the YOYO Claw technology, a terminal side "lobster" AI intelligent agent. This technology relies on intelligent routing algorithms to achieve precise scheduling of end cloud collaboration. 100% end-to-end execution of simple and high-frequency tasks, achieving zero consumption of lexical elements; Complex tasks can be called to the cloud on demand, and efficiency can be improved through context optimization and memory matching, significantly reducing user costs.

In the process of intelligent cockpit evolving towards higher levels, the three major challenges are the soaring demand for computing power, high cloud costs, and complex development. The Sage Box launched by SenseTime Jueying provides a way to crack it. Xiao Feng, CTO (Chief Technology Officer) of SenseTime Jueying, introduced that the device is equipped with a large base model that realizes complex intelligent agent capabilities on the vehicle side, and achieves "zero token cost". According to calculations, a single vehicle can save about 30 yuan in cloud costs per day, and after large-scale deployment, it can save over 100 million yuan per 10000 vehicles per year. The system response time can be shortened to 0.5 seconds.

The development of artificial intelligence cannot be separated from the collaborative empowerment of computing power networks and infrastructure. China Mobile's high-speed data network takes "high-speed circulation, secure endogenous, and compliant traceability" as its core, realizing cross network and cross provincial second level interconnection based on business level, ensuring full traceability of data circulation, greatly improving data transmission efficiency and business response speed.

Liu Hong, Deputy General Manager of China Mobile's Digital Intelligence Division, introduced that the first batch of high-speed data network solutions have been implemented in cities such as Tianjin, Suzhou, and Ganzhou, successfully promoting the construction of key scenarios such as new energy vehicle data fusion applications, banking and financial risk control, efficient circulation of meteorological data, and data flow operation services, and helping to transform data transaction delivery from offline to online.