Baidu Search Moves From Answering Questions to Delivering Tasks

Introduction

Baidu used a product briefing in Beijing to present four major AI upgrades for its mobile app. The announcement goes beyond adding a conversational box to a traditional results page. Baidu is trying to make search understand a user’s broader objective, coordinate several steps, return an organized deliverable and continue monitoring a topic when necessary. The company also introduced an updated search AI API aimed at developers and enterprise users.

What has changed

1. Search is being positioned as a task executor. Baidu says the new experience can conduct several rounds of retrieval and organization for complex questions, then return a checked and summarized conclusion. For recurring tasks such as trip planning or information comparison, it aims to shorten the path from searching to jumping between pages and taking action. With user confirmation, the system can also monitor selected items and send alerts when conditions change. Baidu says the capability now covers more than 170 tool scenarios. Examples presented in the source include combining a child’s grades, location and boarding preferences when comparing schools, and notifying a traveler when hotel prices fall.

2. The AI camera adds continuous interaction. The upgraded camera can identify and track multiple objects at the same time. Instead of describing an object manually, users can tap or draw around it and ask follow-up questions. The image experience can then transition into a live conversation. Baidu also says the feature supports real-time dialogue in nine Chinese dialects, including northeastern, Sichuan, Cantonese and Changsha speech. The intended improvement is not merely better recognition, but a smoother path from seeing something to understanding it.

3. GUI features turn data into visual comparisons. Baidu reports that GUI scenarios on its search pages have expanded fivefold. The company says the system connects to authoritative resources covering government services, public affairs and central-bank information, across more than 20 industries. Visualization components and professional chart templates are designed to make dispersed figures easier to compare and interact with. Baidu is also opening the ecosystem to data providers and other partners.

4. AI content becomes more participatory. The app now includes AI interactive dramas and interactive applications. Users can choose different story paths or operate an application and receive feedback, moving from passive viewing toward participation. Baidu has also launched an AI Future Creation League, offering cash incentives and traffic support to attract creators of interactive content.

The API and the “recursive self-evolution” claim

During the promotion period, Baidu says its search API will be priced as low as 4.9 yuan per 1,000 queries, which it presents as cheaper than the DeepSeek search API. The architecture is described in four layers: a content layer with multilingual indexes, an engine layer built around retrieval, feedback and retraining, an infrastructure layer focused on concurrency and latency, and an application layer supporting API, MCP and Skill integrations.

Baidu characterizes the engine as a multi-agent feedback loop that can improve from real retrieval activity without manual intervention. However, claims such as a more than 50% increase in data trustworthiness and an over 80% improvement in iteration efficiency are company disclosures in the briefing. The source provides no independent test protocol or third-party benchmark, so these figures should be treated as vendor-reported product claims rather than established industry results.

Why it matters

The announcement illustrates a broader shift in search competition. The key question is no longer only whether a system can produce a plausible answer, but whether it can complete a multi-step task reliably. If the approach works, users could receive decision-ready comparisons, executable outcomes and later reminders instead of a collection of links.

The practical value will depend on source quality, agent permissions, user consent and the system’s ability to detect and correct mistakes. Baidu’s release offers a clear product direction, but public examples and independent testing will be needed to assess how stable these capabilities are in everyday use.

Source: QbitAI