DeepSeek R1 May 2025 update. Reasoning model with strong performance on math, code, and complex reasoning tasks.