In a leap for AI, researchers unveil s1-32B, a model that enhances reasoning with a mere 1,000 curated questions. Outshining OpenAI’s o1-preview on math contests, it scores 59.6% on tough science problems using a simple "budget forcing" technique. Fully open-source, it challenges industry norms with efficiency and transparency.

Fei-Fei Li’s s1-32B introduces a new approach to low-cost AI distillation, achieving strong reasoning results at a reported cloud cost under US$50. The model’s efficiency invites comparison with DeepSeek’s US$5.6 million development budget, signalling a shift in how AI capabilities may be scaled.