DeepSeek isn’t releasing a new model this time—it’s rolling out the whole toolkit. On November 13, DeepSeek opened up the capabilities of its app as APIs all at once. According to the company, developers can follow its example to reproduce a complete DeepSeek app 1:1.
What’s being opened up isn’t a new model; it’s still powered by V4.1-Flash (deepseek-flash). What’s changed are the options you can enable in a request: web search, deep-thinking search, file uploads, link reading, and safety checks. All three API interfaces—Messages, Responses, and Chat Completions—support them.
The underlying capabilities have also been split into five independently callable endpoints: search, fetching, OCR, speech recognition, and speech synthesis. Pricing was announced too: search and fetching cost 5 yuan per 1,000 calls, OCR costs 1.5 yuan per 1,000 pages, speech recognition costs 0.39 yuan per hour, and speech synthesis costs 1.5 yuan per 10,000 characters. The two speech services are currently half price for a limited time. The company has also open-sourced a chat demo.
My take: The real significance of this move is that it lowers the barrier to building an AI application by turning model setup, search integration, OCR, and voice configuration into a few calls to ready-made APIs. These scattered capabilities were precisely what small teams used to struggle to piece together. Now DeepSeek is taking apart the entire set of building blocks behind its app and selling them individually. For developers, what they save isn’t just money—it’s also the time it takes to turn an idea into a usable product. Competition in AI applications is shifting from comparing model parameters to seeing who can assemble capabilities into a product fastest.
What’s being opened up isn’t a new model; it’s still powered by V4.1-Flash (deepseek-flash). What’s changed are the options you can enable in a request: web search, deep-thinking search, file uploads, link reading, and safety checks. All three API interfaces—Messages, Responses, and Chat Completions—support them.
The underlying capabilities have also been split into five independently callable endpoints: search, fetching, OCR, speech recognition, and speech synthesis. Pricing was announced too: search and fetching cost 5 yuan per 1,000 calls, OCR costs 1.5 yuan per 1,000 pages, speech recognition costs 0.39 yuan per hour, and speech synthesis costs 1.5 yuan per 10,000 characters. The two speech services are currently half price for a limited time. The company has also open-sourced a chat demo.
My take: The real significance of this move is that it lowers the barrier to building an AI application by turning model setup, search integration, OCR, and voice configuration into a few calls to ready-made APIs. These scattered capabilities were precisely what small teams used to struggle to piece together. Now DeepSeek is taking apart the entire set of building blocks behind its app and selling them individually. For developers, what they save isn’t just money—it’s also the time it takes to turn an idea into a usable product. Competition in AI applications is shifting from comparing model parameters to seeing who can assemble capabilities into a product fastest.
