DeepSeek API Adds App Search, Files, Safety and Voice

DeepSeek has opened the search, file, link-reading, safety and voice features of its consumer app through its developer API. On its API documentation news page, dated 13 November 2026, the company wrote: “The full DeepSeek App experience, now via API.” The same wording appeared in a @deepseek_ai thread at 14:24 GMT on 13 November: “Search, files, links, safety, and voice.”

The release is aimed at developers who want those app functions inside their own products, either as built-in tools on the existing model APIs or as separate endpoints. It is a follow-on from the Search and Fetch APIs DeepSeek opened on 17 October, which TSN covered in its DeepSeek-V4.1-Flash-1017 update. Those two endpoints remain, at the same published price, and are now joined by file parsing, OCR and voice, and by a safety-check switch on the model APIs.

The DeepSeek API now carries the app’s search, files, links, safety and voice tools

The 13 November change-log entry and the news page both say V4.1-Flash, called deepseek-flash, now supports the new capabilities across the Messages, Responses and Chat Completions APIs, “with tool calls and context assembly handled server-side.” Web Search can run on text or images, “with citations in non-thinking mode.” Deep Thinking Search, the company says, can “search and read pages throughout multi-step reasoning.” File Input can parse “PDFs, Office files, images, and more.” URL Reading can “automatically fetch links in user messages to inform answers.” Safety Checks, it says, screen inputs and outputs and return “refusal indicators when risks are detected.”

Search is switched on with deepseek_web_search, and URL reading with deepseek_read_url. Search, Fetch, OCR, ASR and TTS are also offered as standalone APIs. DeepSeek has published an open-source demo chat client that, it says, can import DeepSeek conversation history.

The safety language is more than a product label. The news page says: “Public-facing apps must enable safety checks. Developers remain responsible for content moderation, applicable model registration, and AI-generated content labeling.” That is a requirement plus a residual duty. It does not say DeepSeek’s switch takes the developer’s obligations away.

Tool fees are charged on top of token prices, with a limited-time voice discount

DeepSeek’s pricing page lists the new tools separately from model token charges. The news page is explicit that “extended capabilities incur tool usage fees in addition to standard token charges.” The published tool prices are: Search (text or image) at $0.75 per 1,000 searches; web fetch at $0.75 per 1,000 fetches; file parsing and the OCR API at $0.225 per 1,000 pages; text-to-speech at a list price of $0.45 per 10,000 characters, with a limited-time price of $0.225; and speech-to-text at a list price of $0.116 per hour of audio, with a limited-time price of $0.058. The speech table is labelled a 50 per cent discount and a limited-time price. No end date is given.

The $0.75 Search and Fetch figures are the same prices DeepSeek quoted when those endpoints opened on 17 October. What is new on the meter is file and OCR charging, the voice APIs, and the fact that the model APIs can now call these tools as built-in features rather than only as separate endpoints.

This is one more opening of DeepSeek’s own stack to outside developers, after the company open-sourced a Superagent prototype that works inside a GitHub repository and released DeepInfer, the inference engine it says sits behind its own services. Those earlier releases were about running or hosting models. This one is about wiring the app’s tools, and their bills, into the same API.

Opening the app’s tools is not a change in the token price list

The cleanest way to read the 13 November release is as a packaging decision with a new bill attached. Search and Fetch were already public, at $0.75 per 1,000, from 17 October. They have not been cheapened here, and they have not been withdrawn. They have been built into deepseek-flash as server-side tools, and they have been joined by file parsing, OCR, speech synthesis and speech recognition. Anyone costing a public-facing app has to add those tool fees to the token bill, not swap one for the other.

The limited-time TTS and ASR discount is real on the published table, and it is also temporary by the company’s own heading. Because no expiry date is given, the discounted figures should not be treated as the standing list price. The list prices remain $0.45 per 10,000 characters and $0.116 per hour of audio until DeepSeek says otherwise.

The safety-check rule points in the same two directions. Public-facing apps, DeepSeek says, have to turn the check on. Developers, it also says, remain responsible for moderation, for any model registration that applies, and for labelling AI-generated content. That is not a transfer of legal duty onto the API, and it is not a finding about how well the check performs. An earlier TSN report on Fangcun’s safety evaluation of DeepSeek V4.1 Flash-1017 already treated model-level safety claims as something to be measured, not assumed. This release adds a switch and a developer caveat, not a new independent test.

The next official facts worth watching are ordinary ones: whether DeepSeek names an end date for the voice discount, and whether it later says more about how the safety-check requirement is applied on public-facing apps. Until then, the story is an API that now offers the app’s tools, a price list that adds those tools on top of tokens, and a safety rule that still leaves the developer on the hook.