DeepSeek
DeepSeek develops large language models, including open-weight reasoning models, offered through a chat app and an OpenAI-compatible developer API.
About this tool
DeepSeek is an AI company that builds large language models and makes them available in three ways: a chat assistant, a developer API and, for several models, openly published weights. Its DeepSeek-R1 reasoning model drew wide attention in January 2025 because the code and weights were released under the MIT licence together with a technical report.
For developers, the API follows the OpenAI and Anthropic request formats, so existing SDKs and many coding assistants can switch to DeepSeek models by changing configuration. Usage is billed per million tokens from a prepaid balance, and the documentation lists cheaper off-peak hours. Model names and prices change regularly, so the official docs are the reference.
What you can do
Open-weight models
Several DeepSeek models, including R1 and smaller distilled versions, are published with weights that teams can run and fine-tune on their own infrastructure.
Thinking and non-thinking modes
The current API models can answer directly or spend more computation on step-by-step reasoning for harder math, code and analysis tasks.
Compatible API format
Requests use the OpenAI or Anthropic formats, which lets developers reuse existing client libraries and agent tools with little change.
Token-based billing
Costs follow the number of input and output tokens, with off-peak rates at half price according to the official pricing page.
Who is it for?
- A developer points an existing coding assistant at the DeepSeek API to compare cost and quality.
- A research team runs an open-weight DeepSeek model on its own servers for sensitive data.
- A startup prototypes a reasoning-heavy feature using the thinking mode.
Before you choose
Review the data handling terms and the jurisdiction that applies to the hosted chat and API before sending confidential information; self-hosting open weights is an alternative.
Benchmark comparisons published by the vendor are not independent tests, so evaluate the models on your own tasks before committing.
Frequently asked questions
Are DeepSeek models open source?
DeepSeek-R1 and its distilled models were released with weights and code under the MIT licence; check each model’s licence because terms can differ.
Can I use DeepSeek with OpenAI client libraries?
Yes. The API accepts the OpenAI request format, and the documentation also describes an Anthropic-compatible endpoint.
How is the DeepSeek API priced?
It is billed per million input and output tokens from a topped-up balance, with lower off-peak rates; prices are listed on the official docs.
Check the official website for current features, pricing, and terms.
Similar tools
OpenAI API
CodingThe OpenAI API gives developers programmatic access to OpenAI’s GPT models and tools such as the Agents SDK for building assistants, agents and AI features.
Ollama
CodingOllama lets you download and run open AI models on your own computer, or use them in the cloud, with a local API that connects to apps and coding agents.
Jan
CodingJan is a free, open-source desktop app for chatting with AI models that run offline on your computer, with optional connections to cloud models.
Fireworks AI
CodingFireworks AI is an inference and training platform for open models, offering fast serverless APIs, fine-tuning and dedicated GPU deployments.
OpenRouter
CodingOpenRouter gives developers one OpenAI-compatible API and one credit balance to access hundreds of AI models from many providers, with routing and fallbacks.
Anthropic API
CodingThe Anthropic API, part of the Claude Platform, lets developers build applications and agents on Claude models with tools for code execution, computer use and structured output.