Baidu pivots to AI agents with DuMate, Miaoda, and cloud upgrades
Robin Li bets on 'Daily Active Agents' as the new success metric.
Baidu is reorienting its AI strategy around agents. At the Create 2026 conference, CEO Robin Li unveiled a suite of agent-focused products designed to move beyond traditional large language models. The new metric, 'Daily Active Agents' (DAA), signals a shift toward measuring real-world usage rather than model parameters. Among the launches: DuMate, a general-purpose agent for daily tasks; Miaoda, a code-generation agent for developers; and Baidu YiJing, a multi-agent digital human platform that can orchestrate multiple AI agents for complex workflows.
Supporting this agent-first approach, Baidu Cloud announced significant infrastructure upgrades. The new full-stack AI cloud is built around an agent-centric design, featuring a 'Token Factory' that optimizes inference throughput. Key performance gains include KV Cache hit rates that now exceed 90%, dramatically reducing latency for agent interactions. Additionally, Baidu's Kunlun P800 GPU clusters provide high-efficiency computing tailored for agent workloads, enabling faster response times and lower costs. This combination of new products and cloud upgrades positions Baidu to compete in the rapidly growing AI agent ecosystem, aiming to make AI assistants as ubiquitous as mobile apps.
- Baidu proposes 'Daily Active Agents' (DAA) as new success metric replacing raw model benchmarks.
- New products: DuMate (general agent), Miaoda (code generation), Baidu YiJing (multi-agent digital human platform).
- Cloud upgrades: Token Factory for faster inference, KV Cache hit rates >90%, Kunlun P800 GPU clusters.
Why It Matters
Baidu's agent pivot signals a major industry shift from model size to real-world AI utility and engagement.