Chinese Banks Disclose Daily Token Use in First Half 2026 Reports

As large language models deepen their application in the financial sector, token consumption—a core metric for measuring the scale of AI deployment—is becoming a frequent term in bank earnings reports. Several listed banks disclosed their token consumption scale for the first time in their 2026 interim reports, with some banks seeing daily average token usage grow 18-fold from the end of 2025, and some reaching 33 billion tokens per day. The growth in token consumption directly reflects that the banking industry's large language models have moved beyond the pilot stage and are now fully integrated into application scenarios such as customer service, risk control and compliance, and business review. In March 2026, the National Data Administration officially named tokens as "ciyuan" (word elements), establishing them as a new path for releasing the value of data elements, making the input-output ratio of AI services quantifiable and auditable.

Ping An Bank disclosed in its interim report that its daily average token consumption exceeded 5.3 billion in the first half of 2026, up approximately 130% year on year, with more than 450 large model application scenarios cumulatively deployed. China Merchants Bank's interim report showed that its daily average token throughput in the first half grew over 78% compared with 2025, with 256 domain models deployed, up 40% from the end of 2025. According to Zhou Tianhong, chief information officer of China Merchants Bank, at a shareholder meeting at the end of June 2026, the bank's daily token consumption had reached 33 billion. Zhao Guide, vice president of ICBC, noted at the interim results briefing that the "ICBC Zhiyong" large model had been deployed in over 600 scenarios across multiple fields. Postal Savings Bank of China's interim report showed that it had deployed more than 370 large model application scenarios, with daily interactive token consumption exceeding 10 billion. Bank of Jiangsu deployed 15 models ranging from billion-parameter to trillion-parameter scale in the first half, with daily average token usage growing 18-fold from the end of 2025.

In terms of efficiency gains, China Merchants Bank disclosed in its interim report that AI-driven efficiency improvements contributed 13.88 million equivalent man-hours, with 1,386 intelligent scenarios deployed, up 62% from the end of 2025. ICBC built an AI-native integrated financial services agent in its private banking division, reducing the time to generate comprehensive and personalized service plans for clients from five days to three hours. In the internal control and compliance area, ICBC developed an "AI Smart Review" agent, which enhanced risk identification capabilities for compliance review staff by 130% in the first half. In operations management, ICBC built a full-field voucher element recognition capability, processing individual transactions 25 times faster than manual handling. Postal Savings Bank applied digital humans in self-service equipment to assist with business review, improving manual work efficiency by 40%. Ping An Bank used AI technology to optimize cross-border business processes, with cross-border business lending efficiency at the end of June improving approximately 30% from the end of 2025. Bank of Communications built a shipping and trade review agent, automating tasks such as document sorting, element review, and report generation, reviewing over 500,000 transactions daily and improving processing efficiency by over 70%.