AI News

DeepSeek Opens Full-Suite App API, Letting Developers Replicate the Complete DeepSeek App

East MoneyUpdated Nov 14 08:52

AI summary

DeepSeek has announced the opening of its full-suite app API. By enabling the corresponding parameters in the existing API, developers can obtain conversational capabilities such as deep-thinking search, file input, link reading, and model safety. The underlying voice-related capabilities have also been opened as independent interfaces, including OCR, ASR, and TTS, and can be called individually or in combination. The V4.1-Flash model (deepseek-flash) now supports these capabilities under the Messages, Responses, and Chat Completions interfaces, with tool calling and context stitching handled automatically by the server. In addition to token fees, extended capabilities are charged separately: text/image search and web scraping are both 5 yuan per thousand calls, text recognition is 1.5 yuan per thousand pages, speech recognition is 0.39 yuan per hour, and speech synthesis is 1.5 yuan per ten thousand characters.

Why read this

AI score 67

The original article lists the specific capabilities and calling prices of the full-suite API, so readers can estimate the cost of integrating search, file parsing, and voice into their own applications.

Topics

Tags

Body · AI translation

Article body

DeepSeek announces the opening of the full-suite application API.

On November 13, The Paper reporters learned that DeepSeek announced the opening of its full-suite application API (interface): by enabling the corresponding parameters in the existing API, users can obtain deep-thinking search, file input, and other conversational features; voice and related underlying capabilities have also been opened as independent interfaces, including OCR (optical character recognition), ASR (automatic speech recognition), TTS (text-to-speech), etc., supporting individual or combined calls. Developers can implement complex conversational features in their own applications, and even replicate a complete DeepSeek App 1:1.

In the conversational service of the DeepSeek App (including the web version), capabilities such as web search, file input, link reading, voice interaction, and security work together with model inference to form the conversational experience users are familiar with. However, previously, if developers wanted to implement similar features in their applications, in addition to integrating the existing DeepSeek API, they often had to expend considerable effort to connect external tools and handle call orchestration. With this opening, the various interactive features of the DeepSeek App are directly integrated into the existing API; after enabling the corresponding parameters, the relevant tool calls and context assembly are automatically completed by the server side.

DeepSeek introduced that currently, the V4.1-Flash model (deepseek-flash) already supports the following capabilities under the Messages, Responses, and Chat Completions interfaces:

Web search: In non-thinking mode, after enabling the deepseek_web_search tool, the model can automatically initiate web or image searches based on the question and answer with citations.

Deep-thinking search: In thinking mode, after enabling the deepseek_web_search tool, the model can autonomously search the web and read links during multi-step reasoning.

File input: Supports uploading files in PDF, Office, image, and other formats, completing plain text extraction or mixed text-image parsing, and then assembling them into the conversation context for the model to answer.

Link reading: After enabling the deepseek_read_url tool, the model can automatically recognize links in user messages, fetch web page content, and answer in combination with the context.

Model security: After enabling the security verification parameter, DeepSeek will provide basic risk control, reviewing model input and output, and returning a refusal identifier when risks are hit. If providing application services externally, this parameter needs to be enabled, and obligations such as content security review, large model registration, and AI labeling must be borne.

It is worth noting that when calling the DeepSeek API to use extended capabilities, in addition to Token fees, fees for the relevant tool calls will also be charged: the price for text/image search is 5 yuan per thousand calls, the price for web page fetching is 5 yuan per thousand calls, the price for text recognition is 1.5 yuan per thousand pages, the price for speech recognition is 0.39 yuan per hour, and the price for speech synthesis is 1.5 yuan per ten thousand characters.

The DeepSeek team stated that interaction between humans and models is still full of infinite possibilities, and DeepSeek's existing conversational products are only a relatively basic form. Now, all the capabilities behind the official product are supported for free invocation and flexible combination through the DeepSeek API. Developers are expected to use this to save repetitive basic work and create more diverse, more refined, and more imaginative products.

From an industry perspective, the significance of the full-suite API is not that DeepSeek has opened a few more features, but that it pushes the focus of competition from "model benchmark scores" to "developer start-up costs." Competition among large model vendors is shifting from single-model capabilities to the completeness of the developer ecosystem.

When developers can directly obtain complete components such as search, file parsing, voice, and security risk control through one set of APIs, the hidden costs of switching model providers will rise, and the migration barriers for building applications around the DeepSeek API will also increase accordingly. In overseas markets, DeepSeek's call volume has already reached a certain scale, and the opening of the full-suite API is also intended to convert this call-volume advantage into stickiness at the application ecosystem level.

(Source: The Paper) Middle of the article Embedded advertisement

Source

East Money