Baidu upgrades its app on four fronts as search shifts from answering to delivering

Baidu used a product briefing in Beijing on October 30 to announce changes in four areas of its flagship app: the search box, the AI camera, GUI visualization, and interactive AI content. The headline claim was that Baidu search now improves through recursive self-evolution, backed by a cut in the promotional minimum price of its Search AI API to 4.9 yuan per thousand queries, which the company says undercuts DeepSeek's search API.

The shift is described as moving from one question, one answer to one task, one deliverable. Complex questions are handled with multi-step lookup and synthesis, returning a verified conclusion rather than links; routine transactions are completed outright, compressing find, jump and operate into a short path; and topics that need ongoing attention can be tracked after user confirmation, with proactive alerts when something changes. Baidu says this covers more than 170 tool scenarios.

Two examples in the QbitAI report show the intended framing. A parent researching school admissions gets a comparison that folds commute, tuition and enrollment policy around previously recorded details such as the child's grades, home address and boarding preference. A user planning a five-day Sanya trip gets an itinerary and budget, plus an alert when hotel prices drop, saving 2,100 yuan across five nights.

The AI camera update targets three failures of conventional photo lookup: not seeing everything, not describing it clearly, and not being able to ask follow-ups. The new version recognizes multiple targets at once and tracks them as they move, so several plants or instruments are identified in one shot and moving animals or fish keep their labels. Users point or circle to ask about a specific item, then continue into live conversation, and the camera supports real-time dialogue in nine dialects including Northeastern Mandarin, Sichuanese, Cantonese and Changsha dialect.

On the GUI side, Baidu says visualization coverage in search results has grown fivefold, drawing on more than five billion authoritative resources including government, public services and central bank data across over 20 industries, with AI-generated charts and templates making complex data comparable and interactive. Baidu also opened this layer to outside data institutions and partners.

For content, the app gains AI interactive dramas and interactive applications where users pick story branches or operate things directly and get feedback, turning passive consumption into participation. A companion Baidu AI Future Creation League offers millions in cash and traffic support to interactive creators.

The low price is tied to what Baidu calls a four-layer AI-native architecture. The content layer spans over 80 languages and a hundred-billion-scale index, with data reliability said to be up more than 50 percent; the engine layer runs a feedback loop where every real query and user response becomes training signal across cooperating agents, which Baidu says lifts iteration efficiency by over 80 percent; the architecture layer emphasizes high concurrency and low latency; and the application layer accepts API, MCP and Skill integrations, with Baidu claiming a 20 percent gain for its Wenxin assistant after connecting.

Taken together, Baidu is trying to change what search is judged on: not whether the answer is right, but whether the task got done. That is also the hardest thing for outsiders to evaluate in an agent-style product, which makes real completion rates across those 170-plus scenarios, the methodology behind the reliability figure, and the durability of a promotional 4.9 yuan price the concrete questions to watch.