Tencent Cloud opens Yuanbao Search to developers at ¥14–26 per 1,000 calls
Tencent Cloud opened Yuanbao Search to developers: Express, Standard and Flagship cost ¥14, ¥20 and ¥26 per 1,000 calls, and Flagship is enterprise-only.
On October 20, 2026, Tencent Cloud opened Yuanbao Search to outside developers. Yuanbao Search (Tencent Yuanbao Search, or wsa) is Tencent’s web search service for large models and agents, and developers can reach it through Skill, MCP or API. The official billing page lists three editions at ¥14, ¥20 and ¥26 per 1,000 calls. Only the flagship edition is limited to enterprise customers.
Key points
- Pricing: Express costs ¥14, Standard ¥20 and Flagship ¥26 per 1,000 calls. A Standard resource pack of 50,000 calls costs ¥950 and is valid for one year.
- Limits and latency: Express and Standard allow 20 queries per second, Flagship 5. Tencent’s FAQ gives an average latency of 600 to 800 milliseconds, with P99 near 1,200 milliseconds.
- Free calls: press reports say developers get 2,000 free calls a month. The billing page, the FAQ and the product page do not mention that quota.
Background
Yuanbao Search did not start with this announcement. ChainCatcher, citing Tencent Cloud Internet Search API records, says the service went live on July 1, 2025. A Kuaike report says the index is at the hundred-billion scale, daily calls run into the hundreds of millions, and more than 50 core businesses now use it, including WorkBuddy, ima, Tencent Video and Tencent’s fintech unit. What changed on October 20 is that outside developers can now buy it by the call.
Other vendors are moving the same way. ChainCatcher reports that Alibaba Cloud launched a standalone internet search API on March 20, 2025. On October 17, 2026, DeepSeek opened its in-house Search and Fetch APIs at ¥5 per 1,000 successful requests. Our entry on DeepSeek’s Search and Fetch APIs covers its pricing and billing rules.
The facts
Yuanbao Search’s three editions differ in price, rate limit and features. The prices and limits come from the billing page and the FAQ. The features come from the product page and the API reference.
| Edition | Per 1,000 calls | QPS ceiling | Main difference | Access |
|---|---|---|---|---|
| Express | ¥14 | 20 | Light ranking model; public web sources only; 5–10 results | Not stated |
| Standard | ¥20 | 20 | Base ranking model; adds authoritative vertical sources, dynamic summaries and domain filters; up to 50 results | Not stated |
| Flagship | ¥26 | 5 | Advanced ranking model; adds sub-link attachments and an authority grade | Enterprise only |
Pay-as-you-go billing settles daily and is invoiced the next day. Failed calls are not billed. Standard resource packs cost less per call. The effective-price column below is our arithmetic on the list prices.
| Standard resource pack (valid one year) | Price (¥) | Effective price per 1,000 calls (¥) |
|---|---|---|
| 50,000 calls | 950 | 19 |
| 1,000,000 calls | 18,500 | 18.5 |
| 10,000,000 calls | 175,000 | 17.5 |
Four latency figures appear, from different sources and measuring different things:
| Figure | Source | What it measures |
|---|---|---|
| 8 ms average underlying index retrieval | Kuaike, Tencent News and Phoenix reports | Index retrieval only |
| Ranking model inference cut from 149 ms to 30 ms | Kuaike and Phoenix reports | Ranking model only |
| “As fast as 300 ms” | Product page | Best case, not an average |
| 600–800 ms average, P99 about 1,200 ms | FAQ | Product latency; method not described |
Both access routes require Query. The MCP endpoint is https://api.wsa.cloud.tencent.com/Mcp. It speaks JSON-RPC 2.0 over Streamable HTTP, exposes one tool called SearchPro, and takes a bearer token in the Authorization header. The REST API is served from wsa.tencentcloudapi.com with Action SearchPro and Version 2025-05-08, and it follows Tencent Cloud’s standard request signing. The MCP page caps Query at 1,500 bytes, about 500 Chinese or English characters. The API reference says Express does not accept Mode or Cnt, and that Deeplinks is flagship-only.
The free quota is the hardest claim to check. A Tencent News item from the outlet Bianniushi puts “2,000 free calls a month in the first year” in its headline. Kuaike and ChainCatcher describe a monthly quota in the body, and Sohu calls it a time-limited benefit. The billing page, the FAQ and the product page are silent. The MCP page says the rules are on an activity page and gives no number. We leave the quota out of every price comparison in this piece.
What others say
Most of the performance numbers for Yuanbao Search appear in media reports of Tencent Cloud’s October 20 announcement. Kuaike, Tencent News and Phoenix all carry the 8 ms figure, and Kuaike also gives a 94% recall rate. Tencent says ima’s recalled sources rose about 10% and Tencent Video’s information coverage grew 20%. Phoenix and Sohu say Tencent’s fintech unit shipped a finance search feature within a week. These are vendor claims, and we found no third-party benchmark.
Of the reports we read, only ChainCatcher gives a launch date and competitor prices. Its DeepSeek figure matches DeepSeek’s own guide. Its Alibaba figures, ¥12 for OpenSearch Lite and ¥18 for the Pro edition after an October 15 cut, carry no unit, and we have not verified them.
The latency numbers are the sharpest conflict. The auto-generated summary at the top of Kuaike’s page gives the average retrieval time as 8 ms and drops the words “underlying index” that its body includes, which reads like an end-to-end wait. The 8 ms and 149-to-30 ms figures appear on none of Tencent’s product, overview, FAQ, billing, API or MCP pages. The FAQ’s 600–800 ms is closer to what a caller actually waits, but the page does not say how it was measured.
Our take
For a team building agents, Yuanbao Search’s pricing is simple arithmetic. Standard at ¥20 per 1,000 calls costs four times DeepSeek’s ¥5. DeepSeek’s guide also lists a Search ceiling of 100 QPS per account, against 20 on Yuanbao Standard. The cheaper service has the higher ceiling. We have not tested result quality, coverage or citation format on either side, so the table is a starting point, not a verdict.
At 100,000 calls a month, the list price comes to ¥1,400 on Express, ¥2,000 on Standard and ¥2,600 on Flagship, against about ¥500 at DeepSeek’s rate. Those figures multiply unit price by volume and leave out engineering time.
The catches matter more than the prices. The free quota appears only in press reports. The 8 ms figure is not what a caller waits. The FAQ says WeChat official-account content is not supported, for copyright reasons. Flagship is enterprise-only. Pricing is in yuan, and we have not checked whether overseas developers can open an account.
How to try Yuanbao Search
The quickest route is MCP. In the Tencent Cloud console, turn on Yuanbao Search and create a service API key. Then add the server below to an MCP client, replacing YOUR_WSA_API_KEY with your key. The field names match the MCP listing:
{
"mcpServers": {
"yuanbao-websearch-mcp": {
"type": "streamableHttp",
"url": "https://api.wsa.cloud.tencent.com/Mcp",
"headers": {
"Authorization": "Bearer YOUR_WSA_API_KEY"
}
}
}
}
Once the server is connected, the model can call the SearchPro tool. Query is the only required parameter.
What to watch
Three things will show whether Yuanbao Search is worth the price. The first is the free-quota rules on Tencent’s activity page. The second is DeepSeek. For users of its API inside Harness, Claude Code and Codex, Search and Fetch calls made through server-side tool calls are unbilled for now. Per-call charges start at 00:00 Beijing time on November 1, 2026. The third is whether anyone publishes an independent latency and quality test.