DeepSeek
Free tierFrontier AI models with world-class reasoning — free to chat, open to build on
Free tier available·All audiences·Powered by DeepSeek·API available·Open source
Key strengths
World-class reasoning performance with DeepSeek-R1Open-source models available for self-hosting and fine-tuningLow-cost API access with OpenAI-compatible endpointsStrong coding capabilities via DeepSeek-Coder seriesMultimodal support including vision-language models (DeepSeek-VL)
Free tier + paid plans
Hangzhou, China
Founded 2023
Self-hostable
No ratings yet
DeepSeek's model family is built on a Mixture-of-Experts (MoE) transformer architecture, enabling high performance at reduced inference cost. The API platform is OpenAI-compatible, allowing developers to swap in DeepSeek models with minimal code changes using familiar chat completion endpoints. The lineup spans general-purpose chat (DeepSeek-V3/V4), advanced chain-of-thought reasoning (DeepSeek-R1), code generation (DeepSeek-Coder V2), math (DeepSeek-Math), and vision-language tasks (DeepSeek-VL). Model weights for most series are publicly released on Hugging Face under permissive licenses, enabling self-hosted deployments and fine-tuning.