DeepSeek releases R1 weights and lowers the cost of advanced reasoning models

Summary

DeepSeek published R1, R1-Zero, a technical report and six distilled models, with the core weights under the MIT licence. Its reinforcement-learning approach, open weights and low API pricing expanded access to reasoning models, while benchmark comparisons at launch were based on DeepSeek's own evaluations. This milestone is included for its lasting influence on AI products, research and adoption. Follow the original source for complete technical details, current availability, licensing and later updates.