DeepSeek
DeepSeek opens its app's web search and file input to the API
DeepSeek made web search, URL reading, file input and safety checks from its app available through the API, with per-use tool fees on top of tokens.

DeepSeek has made the features behind its consumer app available to developers through its API, including web search, URL reading, file input and safety checks. Each tool carries its own fee on top of normal token charges.
The company described the release as “now available” in a Nov. 13 post.
The Tools Run on deepseek-flash
The built-in features work on the deepseek-flash model, served by DeepSeek-V4.1-Flash-1017, across the Messages, Responses and Chat Completions APIs. DeepSeek’s service handles the tool calls and context assembly. The Web Tools guide, read Nov. 14, names only deepseek-flash.
Search and URL Reading Are Declared as Tools
A developer turns on web search by declaring the deepseek_web_search tool with thinking disabled. The model then searches the web or images when a question needs it and answers with citations. With thinking enabled, the same tool lets the model search, open pages and find text on its own over several reasoning steps.
The deepseek_read_url tool covers links. The model detects them in the user message, fetches the pages and answers with them in context. The model decides whether to search or read, and a request that names either tool in tool_choice returns a 400 error.
File Input Offers Two Parsing Modes
Developers can upload PDFs, Office files, images and more. The text_recognition mode gives the model text only. The rich_layout_recognition mode also crops charts, photos and diagrams and hands them to the model at their positions in the text. Neither the original file nor the parsing result can be downloaded.
Public Apps Must Turn On Safety Checks
DeepSeek said public-facing apps must enable the safety-check parameter, which reviews model input and output and returns a refusal marker when it finds a risk. Developers remain responsible for content moderation, applicable model registration and AI-generated content labeling.
Standalone APIs and a Demo Client Come With It
DeepSeek also sells Search, Fetch, OCR, speech to text and text to speech as standalone APIs. It published an open-source demo chat client built on them, deepseek-ai/chat-demo.
Tool Fees Add to Token Charges
Search costs $0.75 per 1,000 searches and web fetch costs $0.75 per 1,000 fetches, according to the pricing page as of Nov. 14. Each is billed per use whether a developer calls it directly or the model calls it through the web tools. File parsing and OCR cost $0.225 per 1,000 pages.
Token prices for deepseek-flash at peak are $0.30 per million input tokens on a cache miss and $1.20 per million output tokens, with off-peak rates at half.
Our View
We think the detail that matters most for developers is that the model decides when a search happens. Declaring the tool hands that choice over, and tool_choice cannot name it.
At the Nov. 14 price, one search or one fetch costs $0.00075. The count is what a developer does not control. A question the model judges to need three searches is billed for three. In deep thinking mode the model also opens pages on its own over several steps, and each fetch is billed per use. The tool bill follows how the questions are phrased, so a budget built on request count will miss it.
Developers of public-facing apps must also turn on the safety checks, and moderation, model registration and content labeling stay with them.
Building on these tools pays off only while the model’s search habits stay modest enough that the per-use fees remain small next to the token bill.
