On November 13, 2026, DeepSeek formally opened its App API suite (全家桶应用 API). After you enable the relevant parameters in the existing API, tool calls and context assembly run on the server. These in-conversation capabilities currently work only on V4.1-Flash (deepseek-flash) under Messages, Responses, and Chat Completions. This is not the first time search or fetching has been available; the point is in-conversation tool completion, plus standalone voice APIs.
What features are supported?
The announcement lists web search, deep-thinking search, file input, URL reading, and model safety. deepseek_web_search can search pages or images with citations, and in thinking mode can search and read links during multi-step reasoning. deepseek_read_url detects and fetches links in the user message. File input accepts PDFs, Office documents, and images, then extracts text or parses mixed layouts. Model safety optionally reviews inputs and outputs and returns a refusal flag when a risk is hit. Public-facing apps must enable it; developers remain responsible for moderation, registration, and AI labeling.
Search, Fetch, OCR (text from images or documents), ASR (speech-to-text), and TTS (text-to-speech) are also available as standalone APIs. Standalone OCR returns results page by page; the file is not stored in the Files API or sent to a chat model. For model answers about a document, use official file parsing or in-chat file input.
Which models and APIs can use them?
In-conversation features are currently limited to V4.1-Flash. Standalone APIs are for your own workflows. An official demo and example code can import chat history. They are official samples, not a production-compliance guarantee or proof that third-party relays already offer them.
What else is billed besides tokens?
Tool usage is billed separately from token fees; the 5-yuan search price and discounted voice rates do not cover everything. Prices below come from the official Chinese pricing page, checked November 14, 2026. Voice discounts are limited-time, with no end date, not a permanent 50% off.
- Search (text and images) and web fetch: 5 yuan per 1,000 calls each, for direct Search/Fetch or built-in web tools. Standalone APIs bill only HTTP 200 results (non-empty Search
results; Fetch must return page content). Failures, empty searches, and timeouts are not billed; in-chat web tools have finer rules. - File parsing and OCR: 1.5 yuan per 1,000 pages; OCR is billed by successfully recognized pages; failures or timeouts are not billed.
- TTS: list price 3 yuan per 10,000 characters, limited-time 50% off at 1.5 yuan. The official unit is characters, not Chinese "words": a Chinese character counts as 2; letters, digits, spaces, and other characters count as 1.
- ASR: list price 0.78 yuan per hour of audio, limited-time 50% off at 0.39 yuan, billed to the second; under 1 second rounds up to 1.
If prices change, the official pricing page prevails.
How should developers choose?
For a conversation close to the DeepSeek App, enable the tools in the existing chat APIs and let the server finish tool calls and context assembly. If you only need search, fetch, OCR, or speech, standalone APIs are often cleaner. The bill has two lines: token fees and tool fees.
How we see it
Our view is that this release saves the work of orchestrating tool calls and stitching results into context. It does not make the features free, or available on every model or platform. Add tool fees to token fees at official prices, then choose in-chat integration or standalone calls—more reliable than rushing to recreate the official app.
Sources: official announcement (November 13, 2026); pricing page, checked November 14, 2026.