Reasoning models

12 September 2024

Models trained with reinforcement learning to think at length before answering. OpenAI o1 (September 2024), DeepSeek-R1 (January 2025). Spend compute at inference, not only in training.

Also: Test-time compute