
On 13 November 2026 DeepSeek opened the features of its own chat app to outside developers. Through its existing API, software can now ask the deepseek-flash model to search the web and cite what it found, read the links a user pastes in, answer questions from uploaded PDFs, Office files and images, and run a basic safety review on what goes in and what comes out. For a small or medium business, the DeepSeek API web search and file tools make a research or document chatbot cheaper and quicker to build, and they bring data and responsibility questions with them.
What happened
DeepSeek's API change log entry dated 13 November 2026 is titled "Releasing the full suite of application APIs". It says the web search, file input, URL reading, voice interaction and safety features of the DeepSeek App, including its web version, are now open to developers. A developer switches each feature on with a parameter in the existing API (application programming interface, the door one program uses to ask another for work), and DeepSeek's service runs the tool calls itself.
The features work in the Messages, Responses and Chat Completions API formats. DeepSeek's Build a Chatbot guide says the API now offers the same conversation features as DeepSeek Chat on the web and in the app. DeepSeek has also open-sourced a demo chat client that can import a person's DeepSeek chat history; the guide says it is for demonstration only. Voice input, read-aloud, OCR (reading text from images) and Search and Fetch are also sold as standalone APIs.
The release builds on DeepSeek's update of 17 October 2026, when it opened its Search and Fetch APIs and upgraded the model behind deepseek-flash, which we covered in our brief on the V4.1 Flash 1017 upgrade.
Key details
| Feature | How a developer turns it on | What it does |
|---|---|---|
| Web search | The deepseek_web_search tool, thinking off | Searches the web when the question needs it and answers with citations |
| Deep thinking search | The deepseek_web_search tool, thinking on | Searches the web and reads links on its own over several reasoning steps |
| File input | Upload through the Files API, then refer to the file | Answers from PDFs, Word and PowerPoint files, images and more |
| URL reading | The deepseek_read_url tool | Fetches the pages linked in the user's message and answers from them |
| Safety check | The safety_check parameter set to high | Reviews input and output and returns a refusal when it finds a risk |
The prices on DeepSeek's Models & Pricing page, in US dollars:
- The deepseek-flash model (version DeepSeek-V4.1-Flash-1017): per 1M tokens, $0.15 off-peak and $0.3 peak for input not already in DeepSeek's cache, $0.6 off-peak and $1.2 peak for output. A token is a small piece of text, often part of a word.
- Web search and web fetch: $0.75 per 1,000 uses each, whether the model calls them or your own code does.
- File parsing and the OCR API: $0.225 per 1,000 pages.
- Text to speech: $0.45 per 10,000 characters at list price, $0.225 for a limited time (50% off).
Peak hours are 01:00 to 04:00 and 06:00 to 10:00 UTC, Monday to Friday, excluding Chinese public holidays, and the page says DeepSeek reserves the right to adjust its prices. An uploaded file is kept permanently unless the developer sets an expiry of between 1 hour and 30 days.
Why it matters
A chatbot that looks things up and reads documents used to need several services wired together: a search provider, a page reader, a document parser and a content filter. DeepSeek now offers all of them behind one API key and one bill. For a business that wants a staff assistant to answer from its policy manuals, that is less to build and maintain.
The cost model changes too. DeepSeek's Web Tools guide says the model decides whether to go online, and the searching runs on DeepSeek's service rather than in your software. Each search and page read is billed per use, so the bill depends on how often the model chooses to search. A developer cannot force a search, but can turn all the tools off.
Google moved the other way on price: Gemini 4 Hydrogen costs double the current Gemini 3.8 Flash rate, which is why cost per task is the number to compare.
Responsibility does not move to DeepSeek. Its Model Content Safety guide says the safety check helps you meet compliance requirements but does not replace your obligations as a service provider, and that you bear the primary responsibility for content safety review. The change log adds that anyone offering a service to the public should enable the check and fulfil obligations such as large model registration and AI labeling, without saying which country's rules it means.
Documents behave differently from chat messages. DeepSeek's File Parsing guide says a file is parsed and stored at upload, and is billed if parsing succeeds even when your software disconnects. Neither the original nor the parsed result can be downloaded again, and automatic figure detection can miss some charts and diagrams.
What this means for businesses
You do not need to act if you do not use DeepSeek. If your developers or a software supplier are testing it, or a product you buy is built on it, the new tools raise the usual questions about any AI service that reads your files, plus one about where the data goes.
DeepSeek's privacy policy says it directly collects, processes and stores personal data in the People's Republic of China. If your business is covered by the Privacy Act, Australian Privacy Principle 8 sets the steps you must take before personal information is disclosed overseas. The OAIC's guidance on commercially available AI products says privacy obligations apply to personal information put into an AI system and to output that contains it, and recommends, as best practice, not entering personal information into publicly available AI tools.
If you supply Australian Government agencies, check your contract. The Protective Security Policy Framework lists Direction 001-2025, published 4 February 2025, which requires Australian Government entities to prevent the access, use or installation of DeepSeek products, applications and web services on their systems and devices.
Before anyone builds on the new tools, check:
- What goes in. Decide which documents may be sent, and keep customer records, staff files and health details out unless your privacy obligations allow it. A recruitment office feeding CVs to a chatbot should still confirm candidates' claims itself; a verification platform such as Talent Tabloid is built for that step.
- File lifetime. Set an expiry on every upload and delete files you no longer need.
- The bill. Set a spending limit and watch search counts in the first weeks, since the model decides how often to search.
- Refusals. With the safety check on, an answer can still end in a refusal after it is written, so your chatbot needs a polite fallback message.
- Labelling. The OAIC says public-facing AI tools such as chatbots should be clearly identified as such, and your privacy policy should explain your use of AI.
- Keys. OpenAI's key-safety guidance says never to put an API key in a browser or mobile app and to route requests through your own server; the same applies here, and our post on ASD's advice on protecting AI accounts and API keys covers the wider risk.
- Permissions. Give the chatbot access only to what it needs, as we set out for AI agents in a small business. If peak and off-peak rates are new to you, our post on DeepSeek V4 Pro's pricing explains them.
Whether a DeepSeek-based tool, another provider or no AI at all suits you is a business decision first. Our IT consulting team can help you weigh the data and cost questions, and our custom software and business systems integration work covers building an assistant into the tools your staff already use. Ask us for a free first consultation and a written quote.
Frequently asked questions
What did DeepSeek release on 13 November 2026?
DeepSeek opened the web search, deep thinking search, file input, URL reading and safety check features of its chat app to developers through its API, using the deepseek-flash model. Voice input and read-aloud, OCR and Search and Fetch are available as separate APIs.
How much does DeepSeek's web search cost through the API?
DeepSeek's pricing page lists search and web fetch at $0.75 per 1,000 uses each, billed whether the model calls them or your own code does. Model tokens are charged on top, and DeepSeek says its prices may change.
What does DeepSeek's privacy policy say about where data is stored?
DeepSeek's privacy policy says it directly collects, processes and stores personal data in the People's Republic of China. Its Files API keeps an uploaded file permanently unless you set an expiry of between 1 hour and 30 days.
Does DeepSeek's safety check make a chatbot compliant?
No. DeepSeek's safety guide says the safety check helps you meet compliance requirements but does not replace your obligations as a service provider, and that you bear the primary responsibility for content safety review.
Key takeaways
- On 13 November 2026 DeepSeek opened its app's web search, file input, URL reading and safety check to developers.
- Search and web fetch cost $0.75 per 1,000 uses each, on top of model tokens, and the model decides when to search.
- Uploaded files are kept permanently unless you set an expiry.
- DeepSeek's privacy policy says it stores personal data in the People's Republic of China: check your privacy obligations first.
- The safety check helps, but responsibility for what your chatbot says stays with you.