In early 2025, open reasoning models like DeepSeek R1 popularized explicit chain-of-thought verification. By spending compute time "thinking" before generating responses, reasoning models achieve breakthrough accuracy in complex math, coding, and logic puzzles.