DeepSeek Opens APIs for Search, Speech, OCR and More
Introduction
DeepSeek is opening more of the infrastructure behind its consumer-facing assistant to developers. According to the announcement, functions such as web search, page retrieval, file processing, OCR, automatic speech recognition and text-to-speech can now be accessed through APIs. Developers do not need to subscribe to the entire bundle: the capabilities can be selected and combined according to the needs of a product.
The company says developers could use these interfaces to reproduce an experience close to the DeepSeek app. It has also released a lightweight chat front-end demo that can import historical DeepSeek conversations, giving teams a starting point instead of asking them to build the surrounding product infrastructure from scratch.
What changes for developers
- Modular access: Search, Fetch, OCR, ASR and TTS are exposed as separate services and can be billed independently.
- Less search orchestration: DeepSeek had already released Search and Fetch interfaces, but developers previously had to decide when to search, call the tools, retrieve pages and pass the results back to the model. The new application-level capability lets DeepSeek manage more of that sequence automatically.
- Official speech interfaces: Speech recognition and speech synthesis are now available directly from DeepSeek. Developers building voice assistants no longer have to rely on separate speech providers for these basic functions.
- Broader conversational integration: The announcement says V4.1-Flash supports the related extensions through Messages, Responses and Chat Completions interfaces.
Pricing and access
The source describes a discounted price of 0.39 yuan per hour for speech recognition, reduced from 0.78 yuan. Text-to-speech is listed at 1.5 yuan per 10,000 characters, down from 3 yuan. Search and page fetching remain priced at 5 yuan per 1,000 successful calls, while OCR is listed at 1.5 yuan per 1,000 pages. These figures reflect the pricing and promotion described in the supplied material; developers should verify current terms on DeepSeek’s official pages before deployment.
The significance is not limited to low unit costs. A team building a meeting assistant, for example, could combine ASR with search and summarization. A document product could connect OCR, file handling and question answering in one workflow. Modular interfaces reduce the amount of glue code required before a team can test a product idea.
Why it matters
DeepSeek’s approach suggests that the company is willing to treat its app functions as building blocks rather than keeping them inside one fixed consumer experience. The current default interface for many AI products is still a chat box, but voice interaction, document workflows and search agents may require very different product designs.
Opening APIs does not remove the hard parts. Developers still need to manage authentication, privacy, failure recovery, tool coordination and production costs. Yet the starting point is becoming more accessible. By exposing model-adjacent capabilities as composable services, DeepSeek gives developers more room to experiment with what an AI application can be beyond a conventional chat window.
Source: QbitAI
Comments
Checking sign-in status...
Loading comments...