Service eligibility and regional restrictions
PuppyIP serves only compliant overseas businesses and their authorized personnel. Proxy services are not available in mainland China. The service may only be used for lawful business activities outside mainland China. Use of this service within mainland China is prohibited.
Hosting a proxy IP or server overseas does not change these restrictions. The service must not be provided to end users in mainland China through relaying, forwarding, sharing or resale. Before use, read the Terms of Service.
Key Takeaways
- Identify whether the model triggers a built-in web tool or your application sends a standalone Search or Fetch request.
- Model-triggered results enter the model context and count toward token usage.
- Standalone APIs bill successful requests that return results. The Search result count parameter does not change its per-request rate.
- DeepSeek lists account-level limits of 100 requests per second for standalone Search and 25 for Fetch. Check the intended endpoint when setting concurrency.
What billing changed on November 18?
DeepSeek’s price page sets the effective time at noon on November 18, 2026, Beijing time (UTC+8). The change covers server-side search and fetch triggered by a model call. The returned text enters the context and counts toward model token usage.
For example, if you use the built-in web tools with deepseek-flash to research and answer, track model input, output, and context added by retrieved results. The model-triggered search or fetch does not add a separate per-call charge.
What do standalone Search and Fetch cost?
If your app calls Search to find results or Fetch to read a known page, each API costs $0.75 per 1,000 successful requests. DeepSeek defines a billable success as HTTP 200 with a usable result. Search needs a nonempty result list. Fetch needs page content.
At the published rate, 1,000 successful searches plus 1,000 successful fetches cost $1.50. If you then send the results to a model, budget separately for its token usage.
Which interface fits the task?
When the model needs to research public material and answer within one workflow, start with the deepseek-flash examples in the official Web Tools guide. The November 13 update documents tool support across Messages, Responses, and Chat Completions. Check the current supported model and interface for your integration.
If your application needs structured search results first, or wants to select specific pages to read before downstream processing, consider the standalone APIs. Track Search and Fetch separately. Increasing the number of Search results keeps the same per-request rate and can increase the text sent to a model later.
Check throughput and reconcile usage separately
DeepSeek documents account-level limits of 100 requests per second for standalone Search and 25 for Fetch. Set concurrency for each route, then compare successful request counts against the corresponding billing unit.
Measure both kinds of usage over the same test period: successful standalone calls with results, and tokens used by model calls. If you connect through a third-party cloud endpoint, check that provider’s own features and billing.
Sources
Frequently Asked Questions
Is a standalone fetch billed if it returns HTTP 200 with no page content?
The standalone Fetch guide requires page content in addition to HTTP 200. Search likewise needs a nonempty result list. The guide excludes timeouts and failed requests from these charges.
If I search directly and then send the results to a model, are those separate charges?
Yes. Successful standalone requests and model tokens are billed separately. Apply the selected model’s token rates to the results sent as model input.
How many tool calls can deep-thinking search make?
DeepSeek’s Web Tools guide says the model makes up to about 20 tool calls before answering. Check the chosen model and mode for actual behavior and token usage.