(Yicai) Aug. 21 -- Chinese technology giant Baidu has released a set of major product upgrades, headlined by an entirely free Task Engine 2.0 for its Ernie artificial intelligence chatbot that provides end-to-end autonomous AI agent execution capabilities

Task Engine 2.0 makes the leap to “helping you get things done” from simply “helping you understand,” Baidu said at its AI Day yesterday. The AI agent framework, which is built into Ernie, is designed to manage entire workflows -- from gathering information and processing data to generating content and delivering finished output -- with zero user involvement, it said.

Beyond question and answer interactions, Task Engine 2.0 supports scheduled and condition-triggered push notifications, enabling a new service paradigm of "one-time commission, continuous delivery," Baidu said.

Baidu also announced that every feature will now be available at no cost. Most competing AI agents use subscriber or usage-based payment models, the company noted, adding that by offering comparable capabilities free of charge, Baidu aims to deliver superior cost performance and hasten mainstream adoption of AI services.

Baidu cited benchmarking results to support its performance claims. Last month, the previous version of Task Engine placed first in PinchBench v2, an international benchmark for engineering-oriented AI agents, scoring a 94.6 percent overall success rate and a 94.4 percent average score.

Earlier this month, SuperCLUE -- one of China’s best-known AI evaluators -- ranked Task Engine first in its latest benchmark results for AI crawlers with a total score of 97.62, outperforming Xiaomi's MiMoClaw, Tencent Holdings' Workbuddy, and Moonshot AI's KimiClaw.

Growing user adoption reflects increasing market recognition and user stickiness for Ernie's iterative smart search experience. In June, its daily active users surged 83 percent from a year earlier, while its average daily dialogue rounds more than tripled.

In addition, Baidu Search has launched a new fully interactive full-page search result format, which goes beyond conventional text and image displays by integrating interactive knowledge cards, step-by-step diagrams, interactive animations, and other dynamic components, enabling comprehensive real-time user interaction with search content.

Editor: Martin Kadiev