DeepSeek Models

DeepSeek is a Chinese AI lab's family of open-weight language models. Founded in 2023 as a subsidiary of the quantitative trading firm High-Flyer. DeepSeek gained prominence with V2's efficient mixture-of-experts architecture, V3's GPT-4o-level performance at dramatically lower training cost, and R1's open-weight reasoning capabilities that rivaled o1. All models are released under permissive MIT licenses.