DeepSeek Opens Its App Tools to API Developers: Web Search, Files and Voice — With Prices
DeepSeek has opened the tools behind its chat app to developers. In a change log entry dated 13 November 2026, the Chinese AI lab said the features that make its app work — web search, file input, URL reading, voice interaction and safety checks — are now available through its API, with the tool calls and context assembly handled on DeepSeek's side. Developers switch them on with parameters in existing requests, and the model does the rest. The move turns DeepSeek from a cheap model provider into something closer to a full chatbot platform — and, unusually for this kind of launch, the prices are public from day one.
Application APIs are the ready-made features inside DeepSeek's own chat app — web search, file reading, voice — now switched on inside the developer API, so builders no longer have to wire up those tools themselves. Until now, a developer who wanted a DeepSeek chatbot that could search the web or read a PDF had to build those pieces and bolt them on. Now DeepSeek runs them.
What developers can now do
The built-in tools work with the deepseek-flash model — currently served by DeepSeek-V4.1-Flash-1017 — inside the Messages, Responses and Chat Completions APIs. There are five of them, and each maps to something app users already know.
Web search is the headline. With thinking disabled and the deepseek_web_search tool enabled, the model runs web or image searches as the question requires and answers with citations. With thinking enabled, the same tool becomes deep thinking search: the model searches the web and reads links on its own over multiple reasoning steps, rather than doing a single lookup and stopping. That is the difference between a chatbot that can check a fact and one that can research a question.
File input accepts PDFs, Office files, images and more. After plain-text extraction or rich-layout parsing, the files are added to the conversation context for the model to answer from — the same document Q&A that app users get when they upload a file to a chat. URL reading, through the deepseek_read_url tool, makes the model detect links in the user's message, fetch the pages and answer with them in context. And the model safety check, enabled with a parameter, reviews the model's input and output and returns a refusal marker when it finds a risk.
Behind the built-in tools sit standalone APIs for the same capabilities: Search and Fetch, OCR, speech recognition (ASR) and text-to-speech (TTS). These can be called on their own, outside the model, so a developer can use DeepSeek's search engine or voice models without using its chat model at all. DeepSeek's in-house search engine, opened as a Search and Fetch API on 17 October, is the same one that powers the app; the company says it handles hundreds of millions of search requests a day.
![]() |
AI-generated illustration (generated 15 Nov 2026)
What it costs
The tools are billed separately from tokens, and the prices are small — deliberately so. DeepSeek's documentation lists the following as of 13 November:
| Tool | Unit | Price |
|---|---|---|
| Search | 1,000 successful requests | $0.75 |
| Fetch (URL reading) | 1,000 successful requests | $0.75 |
| OCR (file parsing) | 1,000 pages | $0.225 |
| ASR (speech to text) | 1 hour of audio | $0.058* |
| TTS (text to speech) | 10,000 characters | $0.225* |
* Limited-time 50% discount: ASR lists at $0.116 per hour and TTS at $0.45 per 10,000 characters. DeepSeek gives no end date for the discount.
Failed searches and empty results are not billed, which matters for agents that search often and miss often. On top of the tool fees sit the token prices: deepseek-flash costs $0.30 per million uncached input tokens and $1.20 per million output tokens at peak hours, with off-peak rates at half that — $0.15 and $0.60. Peak hours are 01:00 to 04:00 and 06:00 to 10:00 UTC, Monday to Friday, excluding Chinese public holidays. The API is pay-as-you-go from a prepaid balance, with no monthly fee and no tiers.
Put together, the arithmetic is striking. A thousand web searches cost 75 cents, and a million input tokens cost 15 to 30 cents. A chatbot that searches, reads and answers can now be built on DeepSeek for a fraction of what the same features cost from US providers — which is precisely why this launch matters. The trade-off, as ever with DeepSeek, is the one buyers have to weigh for themselves: the price is public, but so is the US order removing DeepSeek from federal systems, and the data-location questions that come with a China-based provider.
The demo client and the safety question
DeepSeek also published an open-source demo chat client on GitHub, built on these APIs. It runs in the browser, stores the API key and conversations locally, and can import chat history exported from DeepSeek's own app. DeepSeek calls it "for demonstration only" — a starting point for developers, not a finished product.
The safety check deserves a closer look, because it cuts both ways. A built-in content screen that developers can switch on is genuinely useful, and DeepSeek is explicit that public-facing services should use it. But the check runs on DeepSeek's servers, under rules DeepSeek sets, and the company tells developers they remain responsible for their own content review, registration and labelling. Nobody outside DeepSeek has yet tested how the check behaves on sensitive topics. TTS supports 10 languages, including English, Japanese and Korean, while ASR lists 21 — and the limited-time halving of voice prices looks like an attempt to get developers to try them now rather than later.
![]() |
AI-generated illustration (generated 15 Nov 2026)
What this means for developers
The launch builds on two October steps: the 17 October opening of DeepSeek's in-house search engine as an API, and the same day's upgrade of the app and API models to DeepSeek-V4.1-Flash-1017. The direction is clear. DeepSeek started as a model maker that undercut everyone on price. It is now assembling the hosted-tool layer that US providers sell as a platform — search, files, voice, safety — at the same aggressive prices.
For a small team building a chatbot or an agent, that changes the build-versus-buy calculation. The features that used to require three vendors — a model, a search API and a document parser — now come from one, with one bill and one key. The Claude Duet launch showed Anthropic moving in the same direction; DeepSeek's version is cheaper and more openly priced, but with the compliance and data-location questions that come with it.
There is also a competitive read. The DeepSeek and Moonshot distillation report earlier this year showed how closely the Chinese labs watch each other, and Kimi K3.1 Pro shows the Chinese frontier is not standing still either. DeepSeek's advantage has always been price; now it is price plus convenience, and that is a harder combination to beat.
Two questions remain open. The first is how long the voice discounts last — DeepSeek gives no end date, and a price that doubles without warning is a budgeting risk. The second is how the server-side tools perform at scale, under real agent workloads rather than demos. Neither should stop a developer from trying the APIs, but both should be in the plan before a product depends on them.
![]() |
AI-generated illustration (generated 15 Nov 2026)
Key Takeaways
- On 13 November 2026, DeepSeek opened its app's tools to developers via the API: web search, deep thinking search, file input, URL reading and a safety check, all for the deepseek-flash model.
- Search and Fetch cost $0.75 per 1,000 successful requests; OCR is $0.225 per 1,000 pages; ASR is $0.058 per hour and TTS $0.225 per 10,000 characters — both at a limited-time 50% discount with no end date given.
- Tool fees sit on top of token prices: deepseek-flash is $0.30 input / $1.20 output per million tokens at peak, half that off-peak. Failed or empty searches are not billed.
- Search, Fetch, OCR, ASR and TTS are also sold as standalone APIs, and DeepSeek open-sourced a demo chat client that imports DeepSeek chat history.
- DeepSeek says public-facing services should enable the safety check and handle their own content review, registration and AI labelling.
Conclusion
DeepSeek's 13 November launch is a small announcement with a large implication: the cheapest frontier-class model now comes with the tools that used to cost extra, at prices that are public and tiny. For developers, the practical step is simple — try the built-in tools with deepseek-flash, cost a real workload with the price table above, and test the safety check before going public. The open questions are the voice discount's end date and how the tools hold up at scale. But the direction is unmistakable: DeepSeek is no longer just selling a model.
Sources
- DeepSeek API Docs — Change Log, 13 November 2026 ("Releasing the full suite of application APIs")
- AI Newsway — DeepSeek Opens Its App Tools to API Users (13 November 2026)
- DataStudios — DeepSeek Subscription in 2026: Free App, No Paid Plan, API From $0.15 per 1M Tokens
- Toolcolumn — DeepSeek: Free App, API Prices and Off-Peak Hours



Comments
Post a Comment