DeepSeek said in its API change log on 13 November that developers can now build its chat app's main features into their own products. A developer turns each feature on with a setting in an API call they already make. DeepSeek's servers then run the tool calls and put together the context. The deepseek-flash model covers web search, a deeper search mode, file input, link reading and a safety check. Search, Fetch, OCR (reading text from images), ASR (speech to text) and TTS (text to speech) also work as stand-alone APIs.
Before, a developer ran the search, pulled text from files and built the prompt in their own code. Now DeepSeek does that loop on its side. In deeper search, DeepSeek said, the model searches and reads links by itself over several reasoning steps. The safety check reviews input and output and returns a refusal marker when it finds a risk. Pandaily reported that tokens and tool use are billed separately: $0.75 per 1,000 calls for search, the same for web fetch, and $0.225 per 1,000 pages for file parsing and OCR. The sources do not say how many calls one deep search makes.
When the vendor runs the tool loop, cost and data flow move from the developer's code to DeepSeek's meter and servers. A budget owner can ask three things. How many billed calls does one deep search use? Which files and links leave the company? What would it take to switch vendors?