
Screenshots show the official pages on the capture date and may differ from the pages at publication. Select an image to enlarge it.
On November 13, 2026, the Chinese AI company DeepSeek released the features of its chat app (DeepSeek App) as a set of APIs for developers. DeepSeek describes it as "the full DeepSeek App experience, now via API." Features such as Web search, which developers previously had to build themselves, can now be used just by adding a few settings.
Features available with V4.1-Flash
The supported model is V4.1-Flash, released in October (its model name in the API is "deepseek-flash"). The following features are now available.
- Web search: searches the Web and images as needed and answers with sources
- Search while thinking: with the deep-thinking setting, it repeats actions such as searching, opening pages and looking within pages up to about 20 times before answering
- URL reading: reads the pages linked in a question and answers based on their content (up to 10 at a time)
- File input: reads PDFs, Word, PowerPoint and images and uses their content
- Safety checks: reviews input and output and stops the answer if there is a risk
Searching and loading pages happen on DeepSeek's servers, so developers do not need to provide their own tools. They work with any of the OpenAI-style APIs (Chat Completions, Responses) and the Anthropic-style API (Messages).
Standalone APIs and pricing
Search, page fetching, OCR (reading text in documents), speech recognition and speech synthesis can also be used as standalone APIs. They are charged separately from model usage fees.
- Search: $0.75 per 1,000 searches (searches with no results are free)
- Page fetching: $0.75 per 1,000 pages
- File parsing and OCR: $0.225 per 1,000 pages
- Speech synthesis: $0.45 per 10,000 characters (half price, $0.225, for a limited time)
- Speech recognition: $0.116 per hour of audio (half price, $0.058, for a limited time)
Speech synthesis supports 10 languages and speech recognition supports 21 languages, both including Japanese, and the language is detected automatically. The price of deepseek-flash itself is $0.15 per 1M input tokens and $0.6 per 1M output tokens during off-peak hours.
Safety checks are required for public services
DeepSeek says safety checks must be turned on in apps that are open to the general public. It also states clearly that developers remain responsible for content review, model registrations required in each country, labeling AI-generated content, and similar obligations. Tokens used for safety checks are also billed.
DeepSeek also published a demo chat interface that combines these features, along with its source code, on GitHub. API keys and conversation history are stored only in the user's browser.
CoAI's take
APIs with built-in Web search and URL reading are already offered by US companies such as OpenAI and Anthropic. By combining these features with low model prices and offering search and voice in one package, DeepSeek has made it possible to build a chat service using only its own APIs. Support for Japanese speech synthesis and speech recognition also makes it convenient for developers in Japan.
On the other hand, search uses DeepSeek's own search engine, so you will need to test how accurately it finds Japanese-language pages. For business use, it is also worth checking how data is handled and how Chinese regulations apply.