The most valuable asset in AI search may be the plumbing rather than the answer.
Tencent upgraded its Yuanbao search on 20 October and opened it to individual developers and enterprises through Tencent Cloud as an API. Behind it sits an index the company describes as reaching a hundred billion scale, built from public web resources plus Tencent’s own content, and stress-tested inside Tencent products including WorkBuddy and ima.
The strategic move is not the search capability. It is that Tencent is turning a search stack validated inside its own business lines into a general-purpose service external developers can call.
The problem being sold
Two failure modes define AI applications at the moment: hallucination and stale knowledge. A capable model can still invent a fact, and a frequently updated model can still be wrong about yesterday.
That matters more once agents enter the picture. An agent calls tools, and an outdated or incorrect retrieval result propagates through every subsequent call, corrupting task execution rather than one paragraph of output.
Retrieval-augmented generation is the standard mitigation, but self-hosting RAG means building and maintaining source collection, data cleaning, indexing and monitoring. That is a standing engineering cost most application teams cannot justify, which is the gap commercial search APIs are built to fill.
Three ways in, and a free tier
Yuanbao search now offers Skill, MCP and API access routes, which covers agent frameworks, model-context integrations and direct HTTP calls without forcing a single architecture.
The pricing hook is a limited free quota of 2,000 calls a month, rising to 6,000 a month through October and November for developers holding an invitation code, so teams can test before committing.
The trade-off is explicit and unchanged by the packaging. Building retrieval yourself gives control over logic and sources at the cost of continuous engineering. Buying it removes that overhead and shifts dependency for quality and stability onto the provider.
Who this actually competes with
Tencent is not selling a better chatbot. It is selling the retrieval layer that other people’s agents will run on, in a market where the alternative is either expensive in-house infrastructure or a foreign API with its own compliance and latency profile for Chinese deployments.
That is a position with real defensibility, because a hundred billion document index is not something a startup reproduces in a quarter. It also locks in a dependency that is harder to reverse than picking a model vendor.
Search was once the product. It is now the substrate, and whoever owns the index sets the terms for everything built on top of it.
Editor’s note: This is an adapted translation of the original Zhidx report. It has been trimmed and restructured for readability for an international business audience. The full original (in Chinese) is at https://www.zhidx.com/p/603034.html.