tech
March 26, 2026
Researchers trained an OpenAI rival in half an hour for less than $50
Posts from this topic will be added to your daily email digest and your homepage feed.

TL;DR
- A new AI reasoning model named s1 was developed in 26 minutes for under $50.
- The model was created using distillation from Google's Gemini reasoning model and is based on Alibaba's open-source Qwen2.5.
- Researchers found that a small dataset of 1,000 questions was sufficient for effective refinement.
- s1 utilizes test-time scaling to enhance its reasoning process.
- The model reportedly outperforms OpenAI's o1-preview on competition math questions by up to 27%.
- The emergence of such models could disrupt the AI industry by reducing the need for massive expenditures on AI development.
Continue reading the original article