2025H2，中国企业级大模型调用进入显著放量阶段。企业级日均调用量由2025H1的10.2万亿tokens提升至2025H2的37.0万亿tokens，增长263%。千问以32.1%的调用量占比位居第一，豆包与DeepSeek分列其后。
AI Agent成为企业级调用增长的核心驱动力。随着Multi-Agent与Text-to-Action等架构嵌入业务工作流，调用从人发问、模型回答转向按任务编排触发检索、调用工具并执行动作，再将结果回写到系统与流程中，推动tokens消耗持续上行。
研究基于系统性问卷调研形成结论与洞察，面向企业IT负责人、技术管理者及AI项目负责人回收问卷1,000份，其中有效问卷870份，覆盖互联网、金融、消费电子、汽车等重点行业，并覆盖不同营收层级与AI投入规模的企业样本。
In 2025H2, GenAI usage in Chinese enterprises entered a clear scale-up phase. Average daily enterprise token usage increased from 10.2 trillion tokens in 2025H1 to 37.0 trillion tokens in 2025H2, representing a 263% increase. Qwen ranked first with a 32.1% share of usage, followed by Doubao and DeepSeek.
AI Agents are identified as the primary catalyst for enterprise call growth. As architectures such as multi-agent workflows and text-to-action are embedded into business processes, usage shifts from one-off Q&A to task-orchestrated retrieval, tool calling, and action execution, with outputs written back into enterprise systems and workflows. This transition drives sustained token consumption.
Insights are based on a structured survey of enterprise IT leaders, technology managers, and AI program owners. The study collected 1,000 responses and validated 870, covering key industries such as internet, financial services, consumer electronics, and automotive, with broad representation across revenue levels and AI investment tiers.
如果您对中国大模型服务市场有进一步的研究需求，请联系我们：
沙利文 袁先生
E-mail: oliver.yuan@frostchina.com
