---
title: "Tencent's New Search Engine Can Find What Google Cannot"
date: 2026-11-13
category: Foundation Models
site: NeuroAI
canonical: https://neuroai.site/a/na-model-yuanbao-search
language: en
---

# Tencent's New Search Engine Can Find What Google Cannot

> Daily AI industry briefing from Brain-Computer Network: verified facts, sourced numbers.

On November 11, its 28th birthday, Tencent quietly opened a new doorway on the web: so.qq.com. Type it into any browser address bar and you land on a search engine. The company also lets users switch their default engine in the latest QQ Browser to the same service. The name is Yuanbao Search, and Tencent Cloud announced it officially the same day.

This is not Tencent's first attempt at search. The company ran Soso years ago and later absorbed Sogou, with the surviving capability settling into QQ Browser and WeChat's built-in search. What is different this time is scope: the entire pipeline, from query understanding through retrieval, ranking and answer generation, has been rebuilt around a large language model.

## The Content Others Cannot Reach

The most interesting change is not technical. Articles inside WeChat Official Accounts and Channels have long been closed off from the open web, so conventional search engines simply cannot index them. Yuanbao Search places that content on the same results page as the rest of the web.

According to Tencent's official statement, the product official Li Ruizhang framed the goal plainly: official account and Channels content should complement the open web and strengthen overall search quality. Official materials state the index covers hundreds of billions of pages, draws on 157 authoritative sources across 13 vertical industries including finance, healthcare, law and travel, and surfaces official-account articles and Channels videos directly in results for trending topics.

## Three Visible Differences

First, the answer comes first. The results page opens with an AI-generated summary, and a follow-up "AI mode" accepts typed, spoken or photographed questions while remembering the previous turn. Queries about weather or stock prices return structured cards instead of link lists.

Second, advertising is segregated. Sponsored results sit in a separate module labeled "related ads" that can be collapsed as a block, and the official line is that commercial content does not interfere with organic ranking.

Third, provenance is explicit. Every result carries a source label, official websites get a distinct badge, and the academic channel preserves citations to the original papers.

On the engineering side, Tencent disclosed that recall on long-tail queries exceeds 94 percent, retrieval latency stays in the millisecond range, and unit inference cost has dropped by 90 percent. The infrastructure was not built overnight: on October 20, Tencent Cloud had already opened the same retrieval capability to enterprises through APIs, MCP and Skill interfaces, with consumer and business products sharing one system.

## Honest Shortcomings

Independent testing arrived the same day. According to AI Star's hands-on report, searches for NASA and other overseas institutions repeatedly failed to return results, and some academic-paper links proved unstable. The tester's conclusion was blunt: the product still has visible gaps before it can be called a mature search engine.

Li was direct at the launch briefing, saying the team can return to the search market as a reformer or challenger, while acknowledging that Baidu's brand remains formidable and worth learning from. The team also drew its own boundaries: the product will not evolve into a general AI assistant, advertising will stay restrained, and near-term success is measured by retention rather than active-user scale. Ma Jin, Tencent Cloud vice president and the product's engineering lead, set the longer ambition: making search capability an infrastructure for the AI industry.

One mechanism buried in the official materials deserves attention: recursive self-improvement, in which the system mines real usage feedback, labels problems automatically and retrains without waiting for version releases.

China's search market counts its users in the hundreds of millions. Tencent has picked a moment when every incumbent is being squeezed by large models to re-enter the arena. Whether it can genuinely fix the oldest complaint of the category, that advertising blends into answers, will be visible within a few months of real use.

**Disclaimer: This article is based on public information and is for reference only. It does not constitute any investment advice.**

---

Published by NeuroAI (https://neuroai.site/) — https://neuroai.site/a/na-model-yuanbao-search
Free to quote with attribution and a link to the original.
