本文来源:https://news.qq.com/rain/a/20261119A04E7E00
若有版权问题,请来信告知:13951431913#139.com 或在右方提交评论。
18天短命收费终结:DeepSeek搜索抓取费用并入Token计费
2026-11-19 08:53 发布于 北京 TechWeb官方账号
【TechWeb】DeepSeek官网显示,自北京时间2026年11月18日12:00起,调用模型时触发的服务端搜索、网页抓取不再单独计费,仅按模型 Token 用量收费。这意味着,这项落地仅18天的按次收费机制被官方正式取消,开发者的联网调用成本结构从此变得更简单——一次请求,只看Token消耗。
短命的按次计费:从落地到取消仅18天
这项被取消的收费机制,实际上才刚刚起步。10月17日,DeepSeek随V4.1-Flash-1017版本发布,正式向公众开放基于 自研搜索引擎 的 Search与Fetch接口 ,两个接口定价均为5元/千次成功请求。
当时官方为开发者预留了两周过渡期:DeepSeek Harness 、 Claude Code 、 Codex 中的服务端工具调用暂不计费,11月1日0时起才正式按次收费。
也就是说,模型调用触发的搜索和抓取按次计费只执行了18天,便被官方叫停,相关费用全部并入模型Token计费。
根据官方说明,使用以下接口时会触发服务端搜索:联网工具、 Anthropic API 的 web_search工具 、 Responses API 的 web_search工具 。此次调整后,开发者在使用联网工具时,不再需要为每一次搜索单独买单。
需要注意的是,独立的Search与Fetch API不受影响,仍按5元/千次成功请求计费。换言之,本次调整针对的是模型调用过程中由服务端自动触发的搜索和抓取,而非独立的搜索服务本身。
自研搜索引擎 :千亿级网页索引,日处理数亿次请求
支撑这套联网能力的,是DeepSeek的 自研搜索引擎 。该引擎索引了全球千亿级网页,支持多语言搜索,每天处理数亿次请求,并针对SEO作弊、 GEO操纵 和 低质AIGC内容 进行识别过滤,以保证搜索结果质量。
对开发者而言,取消按次计费最直接的变化是成本结构变得更清晰。过去,一次联网调用可能涉及“模型Token费用+搜索次数费用+抓取次数费用”多重账单;如今,账单上只剩下一项——Token消耗。对于深度集成联网工具的应用来说,这既降低了成本核算的复杂度,也消除了“每次搜索都要掂量一下”的心理负担。
从两周过渡期到18天正式收费再到全面取消,DeepSeek在计费策略上的快速调整,也反映出AI行业在商业化探索中“跟着开发者体验走”的趋势:凡是增加使用摩擦的收费项,都可能被重新审视甚至砍掉。
本文来源:https://news.qq.com/rain/a/20261119A04E7E00
若有版权问题,请来信告知:13951431913#139.com 或在右方提交评论。
The end of a short-lived 18-day fee: DeepSeek search and web scraping fees are now included in token-based billing
Posted on November 19, 2026, 08:53, from the official TechWeb account in Beijing
[TechWeb] The DeepSeek official website shows that, starting at 12:00 Beijing time on November 18, 2026, server-side searches and web scraping triggered when calling the model will no longer be charged separately; instead, fees will be based solely on model token usage. This means that the per-use billing mechanism, which was only in place for 18 days, has been officially discontinued. From now on, the cost structure for developers' online API calls will be simpler — each request will be based solely on token consumption.
The short-lived per-use billing: only 18 days from launch to cancellation
This now-canceled billing mechanism was, in fact, just getting started. On October 17, with the release of version V4.1-Flash-1017, DeepSeek officially opened the Search and Fetch interfaces based on its proprietary search engine to the public. Both interfaces are priced at 5 yuan per thousand successful requests.
At that time, the official team provided developers with a two-week transition period: server-side tool calls in DeepSeek Harness, Claude Code, and Codex were temporarily not charged, and per-use billing officially started at 00:00 on November 1.
In other words, the per-use billing for searches and extractions triggered by model calls was only in effect for 18 days before being halted by the official team, with all related fees consolidated into the model token billing.
According to the official documentation, server-side searches are triggered when using the following interfaces: the internet tool, the web_search tool in the Anthropic API, and the web_search tool in the Responses API. Following this adjustment, developers will no longer need to pay separately for each search when using internet-connected tools.
It should be noted that the independent Search and Fetch APIs are not affected and are still charged at 5 yuan per thousand successful requests. In other words, this adjustment targets searches and fetches automatically triggered by the server during model calls, rather than the independent search service itself.
Proprietary search engine: Hundreds of billions of indexed web pages, processing hundreds of millions of requests per day
This suite of networking capabilities is powered by DeepSeek's proprietary search engine. The engine indexes hundreds of billions of web pages worldwide, supports multilingual search, processes hundreds of millions of requests daily, and identifies and filters out SEO fraud, GEO manipulation, and low-quality AIGC content to ensure the quality of search results.
For developers, the most immediate change resulting from the elimination of per-use billing is a clearer cost structure. In the past, a single internet call might have involved multiple charges, such as "model token fees + search count fees + crawl count fees." Now, there is only one item left on the bill — token consumption. For applications that deeply integrate internet tools, this not only reduces the complexity of cost accounting but also eliminates the mental burden of "having to consider the cost of each search."
From a two-week transition period to 18 days of formal charging and then to complete elimination, DeepSeek's rapid adjustments to its billing strategy also reflect the trend in the AI industry of "following the developer experience" in its commercialization efforts: any charging item that increases friction in usage may be re-examined or even eliminated.
