Deepseek

AI research company developing large language models for reasoning, code generation, and general-purpose conversation. Models span flagship generalists, reasoning-focused variants with integrated thinking, and code-specialized models trained on billions of tokens across hundreds of programming languages.

DeepSeek releases models with open weights under permissive licenses, including MIT, for research and commercial use. The API follows OpenAI compatibility so teams can swap base URLs and reuse existing SDKs. DeepSeek-V3 uses a Mixture-of-Experts design for strong performance at lower compute cost compared to dense models of similar scale.

Key capabilities:

• General-purpose chat and reasoning models with thinking mode for complex analysis
• DeepSeek Coder family trained on 2 trillion code tokens across 300+ languages
• OpenAI-compatible API for drop-in integration with existing tooling
• Web chat, mobile app, and API platform access
• Open weights for several models under MIT or permissive licenses

Teams use DeepSeek for coding assistants, RAG and document Q&A, multilingual content generation, and agentic workflows that combine reasoning with tool use. Developers integrate via the REST API or OpenAI-compatible clients. Open weights support self-hosting and fine-tuning for custom domains.