DeepSeek Opens Its Consumer App Features to Developers: Web Search, File Parsing, Link Reading, and Voice in the API
Starting November 13, DeepSeek's API offers the core building blocks of its consumer application: web search with citations, deep search, file and link parsing, content safety screening, text-to-speech, and speech-to-text. Web search is priced at $0.75 per 1,000 queries—roughly thirteen times cheaper than the equivalent tool in the Claude API. An open-source demo client has also been released.

Until now, building an assistant capable of searching the web, reading a PDF, and replying out loud required piecing together a search engine, a document parser, and separate voice models. Starting November 13, DeepSeek offers these building blocks directly within its API: they are the very components powering its consumer app, and activating them now takes just a few parameters.
What is Coming to the API
Using the deepseek-flash model, developers can toggle the deepseek_web_search tool (web or image search, returning answers with citations), pair it with thinking mode for multi-step deep research, enable deepseek_read_url to let the model read links pasted by users, and upload files (PDF, Word, PowerPoint, images) that the service automatically parses into text. A safety_check parameter routes both inputs and outputs through a moderation filter, returning a refusal flag if a risk is detected. Rounding out the lineup are automated chat title generation, text-to-speech, and speech recognition.
The critical technical detail: DeepSeek’s servers execute the searches and fetch the pages. The client simply displays the returned blocks and sends them back in the next turn. It all works across the three API formats DeepSeek already supports—Anthropic’s (Messages) and OpenAI’s (Responses and Chat Completions)—making it easy to reuse existing SDKs. The tradeoff: as soon as any of these tools are enabled, DeepSeek overrides several parameters such as temperature and top_p, ignoring any custom values sent in the request.
Pricing as the Primary Selling Point
The pricing schedule is the real headline. A search query costs $0.75 per 1,000, the same as fetching a webpage. On the Claude API, Anthropic’s web search tool is billed at $10 per 1,000 searches, plus tokens: for an identical volume of searches, DeepSeek is roughly thirteen times cheaper. The model itself costs $0.30 per million input tokens and $1.20 per million output tokens during peak hours (01:00-04:00 and 06:00-10:00 UTC on weekdays), and half that off-peak. File parsing and OCR cost $0.225 per 1,000 pages, while voice capabilities are half-price for a limited time.
To showcase the full workflow, DeepSeek released a demo client under the MIT license. It replicates the core experience of DeepSeek Chat (search, files, voice, titles, light and dark modes) and can import chat histories exported from the official app. The API key and conversations remain stored locally in the browser; DeepSeek stresses that the project is intended solely as a demonstration.
What You Need to Know Before Adopting It
DeepSeek shifts part of the regulatory burden onto developers. Its documentation instructs anyone launching a public-facing service to turn on the safety check and fulfill obligations such as content security reviews, large model regulatory filings, and AI-generated content labeling “on their own,” without specifying which jurisdictions these rules refer to. The provided filter is explicitly described as a “basic” risk control. Finally, the performance of the proprietary search engine—launched as an API on October 17 and claimed to index hundreds of billions of pages—remains to be independently assessed.
Our take
With this release, DeepSeek is no longer just selling a low-cost model; it is providing the entire stack for an AI assistant: search engine, document parser, safety filter, and voice. For small teams, this is a genuine shortcut, and search pricing that is thirteen times lower than Anthropic’s could fundamentally reshape the economics of autonomous agents querying the web dozens of times per task. Yet two caveats stand out. First, vendor lock-in runs deep: locked hyperparameters, proprietary blocks that must be passed back as-is, and an in-house search engine make switching providers considerably harder than with a vanilla LLM API. Second, the safety guardrail is merely “basic” and compliance is left entirely to the developer; services targeting European users will still need to verify their own legal responsibilities, particularly regarding data transit and residency. We recommend testing it primarily for internal tooling or early prototypes, while keeping a close eye on the quality of cited sources.
Sources
- DeepSeek API Docs — Change Log (November 13, 2026, 'Releasing the full suite of application APIs') Primary source
- DeepSeek API Docs — Build a Chatbot Primary source
- DeepSeek API Docs — Models & Pricing Primary source
- GitHub — deepseek-ai/chat-demo (MIT license) Primary source
- Claude Platform Docs — Web search tool (Web search pricing) Primary source
A primary source comes directly from the company, lab or organization concerned. A secondary source reports the information without being its author: we flag it as such. Spotted an error? See our corrections policy.


