What is DeepSeek and what is it used for?
DeepSeek is an AI lab that offers language models through an API compatible with the OpenAI format and publishes model weights on Hugging Face for anyone who wants to run them on their own.
The official documentation shows the base URL api.deepseek.com for clients in the OpenAI format and a separate address for clients in the Anthropic format. The current API models appear in the documentation's model list, with a context window of about 1 million tokens, according to the models page.
Reasoning goes through a parameter: thinking mode is turned on or off in each request, instead of there being a separate model for reasoning. The old names deepseek-chat and deepseek-reasoner were retired, and DeepSeek itself announced they would become inaccessible after July 24, 2026.
Typical use is generating and analyzing text and code through the API, calling external functions, and extracting answers as JSON. The model cards on Hugging Face show that earlier releases, such as R1, carry an MIT license.
