Artificial intelligence developer DeepSeek released its full suite of application program interfaces on November 13, opening up search, document parsing and voice tools.
The release allows third-party software teams to replicate the core features of the consumer DeepSeek App inside their own commercial storefronts and customer service platforms.
Under the update, the deepseek-flash model, currently running DeepSeek-V4.1-Flash-1017, processes five built-in tools across its Messages, Responses and Chat Completions interfaces. DeepSeek handles tool calls and assembles conversation context directly on its own servers rather than requiring external orchestration code.
The web search tool operates in two distinct operational modes. In standard non-thinking mode, the model queries web pages or image sources when prompted and returns answers containing source citations. When deep thinking mode runs, the model conducts multi-step reasoning, searching queries and reading linked web pages autonomously across intermediate problem-solving steps.
Document handling processes PDFs, Office documents and images. The system extracts raw text or parses layout formatting before feeding the structured contents directly into the active conversation context window.
A dedicated link-reading function fetches web pages supplied inside customer prompts. To manage compliance, DeepSeek included a safety verification parameter that screens prompt inputs and model outputs, returning refusal flags when content breaches risk limits.
Beyond the unified chat completion endpoints, DeepSeek published standalone interfaces for optical character recognition, automated speech recognition and text-to-speech conversion. The company open-sourced a browser-based chat demonstration client that connects directly to the API endpoints and imports existing DeepSeek conversation logs.
Usage fees for these tools run on top of base token costs.
Source link







