DeepSeek stops billing model-triggered web search and fetch calls
Built-in web tools now cost only the tokens they add. The standalone Search and Fetch APIs keep their $0.75 fee.
DeepSeek stopped charging per call for the web searches and page fetches its models run on their own. From 12:00 Beijing time on November 18, 2026 (04:00 UTC), these server-side tool calls are free, and the text they return is billed as ordinary model tokens. Five days earlier, DeepSeek's pricing page listed the same calls at $0.75 per 1,000, on top of token charges (our November 13 coverage).
What changed
- Server-side search (text and image) and web fetch calls that the model triggers are no longer billed as tool calls. Their results are added to the model context and billed at the model's input token rate.
- This covers the web tools (
deepseek_web_search,deepseek_read_url), theweb_searchtool in DeepSeek's Anthropic-format API and theweb_searchtool in its Responses API. DeepSeek's Search & Fetch guide also lists Web Search in Claude Code, and DeepSeek Harness when its web search provider is set to server-side search or Auto (the default). - The standalone Search API and Fetch API, called directly from your code, still cost $0.75 per 1,000 successful requests each. Only HTTP 200 responses with a result are billed.
- Client-side setups that call those APIs stay billed, for example DeepSeek Harness with the provider set to DeepSeek Search API, and Codex Alpha Search.
- DeepSeek's web tools guide says deep thinking search makes up to about 20 tool calls before it answers. At the old rate, 20 calls added $0.015 to a request in tool fees alone. That fee is now zero, but every page the model reads still counts as input tokens.
DeepSeek prices
From DeepSeek's pricing page on November 19, 2026, per million tokens for deepseek-flash (DeepSeek-V4.1-Flash-1017). Off-peak rates are half of peak. Peak hours are 01:00 to 04:00 and 06:00 to 10:00 UTC, Monday to Friday.
| Rate | Off-peak | Peak |
|---|---|---|
| Input, cache hit | $0.003 | $0.006 |
| Input, cache miss | $0.15 | $0.30 |
| Output | $0.60 | $1.20 |
| Tool | Unit | Before Nov 18 | Now |
|---|---|---|---|
| Model-triggered search | 1,000 calls | $0.75 | Not billed |
| Model-triggered web fetch | 1,000 calls | $0.75 | Not billed |
| Search API (standalone) | 1,000 successful requests | $0.75 | $0.75 |
| Fetch API (standalone) | 1,000 successful requests | $0.75 | $0.75 |
DeepSeek V4.1 Flash 1017 on RouterPlus
The built-in tools are part of DeepSeek's own API at api.deepseek.com. For model tokens, the RouterPlus catalog lists DeepSeek V4.1 Flash 1017 as deepseek/deepseek-v4.1-flash-1017 on 9 hosts. Per million tokens, from the catalog on November 19, 2026:
| Provider | Input | Cached input | Output |
|---|---|---|---|
| Sail Research | $0.09 | $0.006 | $0.30 |
| Wafer | $0.13 | $0.008 | $0.50 |
| Bitdeer AI | $0.144 | $0.0029 | $0.576 |
| RunInfra | $0.15 | $0.01 | $0.60 |
| Doubleword | $0.15 | $0.003 | $0.60 |
| Alpha.sh | $0.15 | $0.003 | $0.60 |
| Particle AI | $0.2125 | $0.0255 | $0.85 |
| Munito AI | $0.27 | $0.0054 | $1.00 |
| Nebius | $0.30 | $0.006 | $1.20 |
What to do
If your app pays for search through the standalone Search API, check whether the model can run the search itself. With DeepSeek's web tools, the call is free and you pay only for the tokens the results add. Watch context size, since long fetched pages now drive the cost.
"tools": [
{"type": "deepseek_web_search", "name": "deepseek_web_search"},
{"type": "deepseek_read_url", "name": "deepseek_read_url"}
]