Starting at 12:00 on November 18, DeepSeek officially implemented new API billing rules. Server-side search and web scraping features triggered by developers when calling large models will no longer be charged separately, and billing will now be based entirely on model Token usage. This adjustment further simplifies the commercial billing process for large models and directly lowers the overall cost for developers building internet-connected AI applications.

According to rules disclosed by DeepSeek, the server-side search capability made free this time covers a diverse range of developer ecosystem interfaces. Specific trigger scenarios include its native web access tool, the Anthropic API-compatible web_search tool, and the Responses API web_search tool.
Previously, DeepSeek officially launched the Search and Fetch APIs on October 17 of the same year. Both APIs rely on its self-developed search engine and were billed by the number of successful requests, priced at RMB 5 per 1,000 requests. With the new rules now in effect, the above retrieval and scraping capabilities have been deeply integrated into the model’s regular invocation pipeline and have become a basic standard feature.
As large models accelerate their evolution toward Agents with autonomous retrieval and real-time interaction capabilities, high-frequency access to external information is becoming a core cost item limiting application deployment. By moving search and scraping from standalone value-added billing into the Token consumption system, DeepSeek not only reduces the combined computing and API expenses in multi-round retrieval scenarios, but is also expected to further promote the large-scale adoption of internet-enhanced AI applications in the developer community.




