DeepSeek released the R1 reasoning model alongside a free consumer application in January 2025