DeepSeek-R1-Lite-Preview
Among the first public o1-style reasoning models outside OpenAI: shows its thinking live, with AIME scores rising as thought length grows.
A reasoning model on chat.deepseek.com, claimed at o1-preview level on AIME and MATH with a transparent chain of thought and a published test-time scaling curve. Weights and API were 'coming soon'; neither shipped until R1 two months later.
- Date
- Wednesday, 20 November 2024
- Lab
- DeepSeek
- Kind
- model
- Access
- app only
Claimed o1-preview parity is company-stated (benchmark scores were only in a chart). Alibaba's QwQ-32B-Preview followed on 2024-11-28; precedence among non-OpenAI reasoning previews is contested.
Sources
This record was checked against its sources on 6 October 2026. How we check