Baidu launches the "O Plan": From search to "agent orchestration"
```
Author | Zhou Zhiyu
On the eve of the Spring Festival in 2026, the tension in China’s internet sphere was fiercer than ever before.
On February 10th, Wallstreetcn learned from Baidu insiders that Baidu had recently quietly launched a project codenamed the “O Plan.” Unlike previous interdepartmental collaborations, this time the Search and Cloud teams are working together, breaking down departmental barriers—MEG and ACG are joining forces.
Other insiders revealed that this project is related to the Baidu App. Baidu is targeting users’ daily scenarios, using the Baidu App as a hub, leveraging the Wenxin Assistant, and mobilizing Baidu’s internal ecosystem and partner services to solve real user needs.
This series of moves sends a strong signal. Baidu realizes that the competition for AI entry points is no longer just about front-end UI—it’s a comprehensive battle over computing power costs, inference speed, and data scheduling capabilities.
To put it bluntly, the UI is just a dialog box that anyone can build; the real generational gap is determined by the scheduling logic and computing power ledgers behind the scenes. When a user makes a complex request, the system must instantaneously mobilize services across the entire ecosystem. If scheduling ability is weak, the AI is just a flashy chatbot unable to truly solve problems; if computing efficiency is low, the cost of a single response becomes absurd, making business unsustainable.
The launch of Baidu’s “O Plan” didn’t come out of nowhere. In an internal sharing session at the start of 2026, Baidu founder Robin Li made it clear: the purpose of training models is inference.
This means that inference efficiency will determine the generational competition in search. If Baidu cannot integrate the underlying search and cloud layers to greatly reduce the cost per intent inference, so-called “AI search” would simply become a money-losing proposition.
In the past, the Search team’s trillions of queries were cash cows—a stock business; the foundation model team represents the future—increments. In the past two years, Wenxin Assistant and Baidu Search functioned as parallel logical lines. But in 2026, with the “O Plan,” that status comes to an end.
Baidu Search is no longer a display gallery, but the knowledge base for Wenxin Assistant. Wenxin Assistant is now the one holding the steering wheel.
Wenxin Assistant is the front-end, responsible for sensing user emotions and ambiguous intents; Search is the posterior, responsible for extracting truth validation from massive real-time data.
This model means Baidu no longer obsesses over short-term Search Ad CTR (click-through rate) fluctuations. Because if Baidu does not self-innovate, ByteDance’s Doubao and Alibaba’s Qianwen will disrupt the traditional search ad model with ‘low-ad-interference, direct results’ kinds of interactive experiences.
The Spring Festival of 2026 is the “D-Day” moment for China’s foundation model industry.
With Baidu, Alibaba, Tencent, and ByteDance collectively joining the red envelope wars, the battle has long surpassed mere marketing, evolving into a life-or-death struggle over the “default entry point.”
In the competition for AI entry point, the major players are following entirely different paths based on their DNA.
As a Spring Festival Gala partner, Doubao’s strength lies in immensely powerful content distribution. ByteDance doesn’t teach users how to think; rather, it uses short videos, social graphs, and ultimate UI interaction to make the AI assistant a new pastime for users. Doubao’s high DAU is its core weapon; it seeks to use high-frequency to subsume low-frequency needs, making AI a part of life.
Qianwen’s 3 billion yuan “Spring Festival treat” campaign clearly signals its intent—AI must generate transactions. Alibaba deeply integrates AI with local life services and e-commerce—you can order milk tea or return goods with a single sentence. The essence of Alibaba’s AI entry point is a universal remote controller for its massive business empire.
Yuanbao, backed by WeChat, bets on the close-knit social graph. Yuanbao is in no rush for independence; it operates more like a smart upgrade to the WeChat ecosystem, using red envelope distribution in social networks to internalize AI as a tool for social interaction.
Now, Baidu too has found its positioning. It wants to solve user issues that are defined and perhaps complex, relying on its depth of content accumulated through search and its full-stack AI capabilities.
When Wenxin Assistant handles intent planning, Search handles knowledge validation, and Baidu Intelligent Cloud handles rapid, low-cost, high-concurrency inference execution, the prototype of an AI agent scheduling center may emerge.
By 2026, the so-called “hundred-model war” has completely dissipated. The industry has entered an extremely brutal period of territory segmentation at the entry-point level.
Baidu, relying on search scenarios, holds a natural advantage in the AI entry point race, but also inevitably faces the challenge of independent app traffic diversion.
Search itself is the most natural ground for Q&A. If the “O Plan” successfully transforms search into an agent distribution platform, Baidu may evolve into the carrier of the AI era.
In the past, the discussion around entry points was about traffic. Now, it's about mental occupancy. Whoever enables users—when they have a need—to speak directly to their phones instead of opening up an app to search, will control the future’s pricing power.
Baidu’s “O Plan” is an all-in gamble, betting that users will become sticky to in-depth answers. ByteDance and Alibaba, in contrast, are betting on delight and convenience. This differentiated showdown will play out in every red envelope exchange and every voice interaction over the holiday.
This isn’t just a competition over technology; it’s a game of courage. At the AI entry point, there is no retreat.
Risk Warning and DisclaimerThe market has risks; investment must be cautious. This article does not constitute personal investment advice, nor does it account for any particular user's specific investment objectives, financial situation, or needs. Users should consider whether any opinions, viewpoints, or conclusions in this article fit their circumstances. If investing based on this, you bear responsibility for the results. ```