StorySoftware
DeepSeek previews R1-Lite, its first reasoning model
DeepSeek previewed R1-Lite, its first reasoning model, on its chat site, showing step-by-step thinking in real time. DeepSeek said it matched OpenAI's o1-preview on AIME and MATH benchmarks and that scores rose the longer it was allowed to think; open-source models and an API were promised later.
- R1-Lite-Preview available on chat.deepseek.com; shows step-by-step reasoning in real time
- Performance: matches o1-preview on AIME and MATH benchmarks
- Inference scaling: accuracy improves as thinking time extends on challenging problems
- Open-source models and API access promised later