DeepSeek announced the launch of a complete set of application-layer APIs on November 13. Previously, developers experienced web search, document input, link reading, voice interaction, and security verification features within the DeepSeek application, all integrated with model inference. To implement similar functionalities in their own products, developers often needed to invest significant effort in integrating external tools and handling API calls in addition to accessing the existing DeepSeek API. Now, these capabilities have been consolidated into a single interface.
DeepSeek's approach involves enabling relevant parameters in the existing API, with tool calls and context stitching automatically handled by the server. Currently, the V4.1-Flash model (deepseek-flash) supports these capabilities under the Messages, Responses, and Chat Completions interfaces: in non-thinking mode, enabling deepseek_web_search allows the model to automatically initiate web or image searches based on questions and provide citations in responses; in thinking mode, enabling the same tool allows the model to autonomously conduct web searches and read links during multi-step reasoning; document input supports uploading formats such as PDF, Office, and images,
which are then integrated into the dialogue context after completing plain text extraction or mixed text-image parsing; after enabling deepseek_read_url, the model can automatically identify links in user messages, fetch webpage content, and respond in conjunction with the context.
DeepSeek API Integrates Multiple Features to Enhance Developer Experience
In terms of security, after enabling verification parameters, the server will provide basic risk control, reviewing the model's inputs and outputs, and returning a rejection indicator if risks are detected. DeepSeek also reminds users that if they provide application services externally, they must enable this parameter and assume responsibilities for content safety review, large model registration, AI identification, and other obligations.
To accommodate developers' existing workflows, the aforementioned underlying capabilities and voice functionalities can also be called as independent interfaces, including Search, Fetch, OCR, ASR, and TTS. With these interfaces, developers can achieve functionalities that are essentially consistent with the DeepSeek application. To visually demonstrate the effects, the official team has also built and open-sourced a lightweight front-end demo based on the new API, supporting direct import of DeepSeek's historical dialogues.
Regarding pricing, in addition to token fees, there will also be charges for the use of the corresponding tools when invoking extended capabilities.