Baidu triples revenue target for its office agent

Baidu Dazi, an AI agent built for office work, surprised its own creators. Jiemian News reported that ARR for the corporate version grew several times over month on month in July and August. Management revised the annual revenue target more than once. The latest figure is three to five times higher than the plan set at the start of the year.

The product launched in March 2026 and comes in two subscription tiers, personal and corporate. AI Newsletter's July 2026 ranking put monthly active users at 6.74 million, up 1,064 percent month over month. That is second place among office agents and first in growth rate.

Product director Li Jingqiu describes a clear split in demand. Technology, finance and manufacturing companies bring in the most revenue. Energy, public administration and transport show strong willingness to buy and are growing fastest this year. Education and retail chains have plenty of use cases, but their purchases depend on a measurable return on investment.

What Baidu is not doing right now is more interesting. People the outlet spoke to say management is not pushing hard on revenue targets or user growth. It cares instead about the agent's execution layer and whether a multi-step task gets carried through to the end. Internally, Baidu positions Dazi as the entry point to its services in the agent era. It integrates search, maps, the encyclopedia and analytics tools, and is meant eventually to connect outside developers' tools as well.

Competition is not standing still. Within a year Alibaba, ByteDance and Tencent all launched their own office products, and their interfaces look increasingly alike. Competition is therefore moving beyond the product itself, to how well complex tasks are executed and to the ecosystem around the platform. Alibaba opened a developer platform for its Workbuddy in September, letting companies bring in their own systems and industry skills.

DeepSeek offers a useful comparison. It monetises its model through a subscription for developers, and after it raised API prices its annual revenue passed a billion dollars. Baidu is taking a different route, selling a subscription for a ready-made workstation rather than access to tokens.

The difference between the subscription model and per-token billing is fundamental. With tokens, revenue grows with the number of calls, but so does the cost of serving responses, and the client sees a bill every time its employees use the tool. With a subscription, the provider carries the usage risk and sells an outcome, not a resource. That is more convenient for the buyer and harder for the seller, because it requires predicting how much work the agent will actually do. The high pace of ARR growth shows that for now corporate clients are leaning towards the second model.