DeepSeek is an AI research company that develops large language models (LLMs), programs that read a prompt and write an answer. It publishes models such as DeepSeek-V4, DeepSeek-R1 and DeepSeek-Coder as open source. People use the models through a chat website, a mobile app and a desktop program called DeepSeek Harness; other programs use them through an API, an interface that a program calls over the internet.
The API accepts requests in the OpenAI and Anthropic formats, so software written for those services can switch by changing the base address and key. In October 2026 it served two models, deepseek-flash and deepseek-v4-pro, each with a context of 1 million tokens, a thinking mode and a non-thinking mode, JSON output and tool calls.
The API is pay per use. The price is set per million input and output tokens (a token is the smallest unit of text that the model reads, such as a word, a number or a punctuation mark), and off-peak hours cost half.