DAA Explained: Li Yanhong’s New AI Metric That Outperforms Token

DAA: What It Is and Why Li Yanhong Says It’s a Better AI metric Than token

On May 13, 2026, at the Baidu Create 2026 AI Developer Conference, Baidu founder Li Yanhong introduced a new concept: DAA, or Daily Active Agents.

According to Li, Token only represents cost, not value. It measures input, not ouTPUt. The true metric for the AI era, he argued, is how many Agents are ACTively delivering results for humans.

The proposal has sparked debate — some say Baidu is reDeFining metrics to mask its weaknesses, while others see it as a forward-looking industry shift.

What’s Wrong with Token?
Token is the smallest unit of text an AI model processes. It’s currently the primary billing and measurement unit for AI services. However, Token tracks computational cost, not actual value delivered. Asking an AI to write a useful email or recite a poem 100 times may consume the Same number of Tokens, but the outcomes are vastly different.

From DAU to DAA
In the mobile internet era, DAU (Daily Active Users) was the core metric for platform success. DAA APPlies the same logic to AI Agents: How many agents are working daily and delivering results?

Li predicts global DAA will easily exceed 10 billion, surpassing the ~3.4 billion DAU of today’s top platforms like Meta. One company could run hundreds or thousands of agents simultaneously.

Is DAA Better Than Token?
While Li’s proposal aligns with Baidu’s strategic positioning — where agent applications are a strength — the concept holds economic merit. DAA focuses on value delivery (output), whereas Token focuses on resource consumption (input). High Token usage with poor task completion signals inefficiency, not prosperity. As Autonomous Agents become measurable, DAA is increasingly practical.

When to Use Token vs. DAA

• Token: Best for evaluating model training cost, inference efficiency, and infrastructure usage — a cost-side metric for cloud and chip providers.

• DAA: Best for measuring platform vitality, application-layer success, and real user value — an output-side metric for valuing AI application companies.

The two metrics are complementary, much like MAU and GMV in the mobile era.

Why It Matters
Li’s proposal signals a shift in AI industry focus — from model-centric competition to application-layer value. The number and quality of active agents will define the next phase of AI leadership, moving beyond raw parameter counts toward real economic impact.