Ping An CTO: Daily Token Use Surpasses 300 Billion as Enterprise AI Scales Up

At the 2026 Apsara Conference held in Hangzhou on September 22, Ping An Group Chief Technology Officer Wang Xiaohang delivered a keynote titled “From T to Value: The Ping An Approach,” outlining the company's AI deployments in finance and healthcare as well as its collaboration with Alibaba Cloud.

According to Wang, over the past year Ping An's daily token consumption jumped from about 30 billion to more than 300 billion. During the same period, computing power rose from 800 PFLOPS to 1,500 PFLOPS, meaning less than a 2x increase in compute supported a 10x growth in token usage. Through full-stack optimization of models, inference and compute scheduling, token output per unit of compute increased by 368%, while the cost per token dropped by 76%.

Wang noted that for large enterprises, scaling AI is not simply about adding more compute; improving token efficiency is key to turning general-purpose intelligence into domain-specific capability that can be embedded in real business processes.

The company said AI has been applied across customer service, healthcare and software development. Serving more than 250 million customers, Ping An's AI quick-service feature has handled over 100 million interactions within six months of launch. AI doctors now handle initial triage for 25% of in-hospital services and 70% of home-based services. In R&D, about 80% of new code is generated by AI, and AI-based testing accounts for 67%.

Ping An and Alibaba Cloud have also been working together around the Qwen family of models to adapt general-purpose capabilities into specialized financial and medical AI. Ping An's financial large model topped the CNFinBench public leaderboard, while its medical large model achieved the highest score on HealthBenchHard, with hallucination rates below 0.3% in clinical questioning scenarios.