Researchers created an open rival to OpenAI's o1 'reasoning' model for under $50

Researchers from Stanford and the University of Washington have developed an open-source AI reasoning model named s1, which can be trained for under $50 in cloud computing credits. This model demonstrates performance comparable to advanced models like OpenAI's o1 and DeepSeek's R1 in math and coding assessments. The s1 model was created by fine-tuning an existing base model through a process called distillation, which extracts reasoning capabilities from another AI model, specifically Google's Gemini 2.0 Flash Thinking Experimental. The researchers utilized a small dataset of 1,000 curated questions and answers, achieving strong benchmark results after a training period of less than 30 minutes using 16 Nvidia H100 GPUs. The development of s1 raises concerns about the commoditization of AI models, as it suggests that significant innovations can be achieved with minimal investment. While major AI labs express dissatisfaction with this trend, the researchers emphasize that their work aims to simplify the process of achieving strong reasoning performance. The paper indicates that while distillation is effective for replicating existing capabilities, it does not necessarily lead to the creation of significantly superior AI models.

- Researchers developed the s1 model for under $50, rivaling advanced AI models.

- The model was trained using a small dataset and a distillation process.

- s1 achieved strong performance in math and coding benchmarks.

- The development raises questions about the commoditization of AI technology.

- Distillation allows for cost-effective replication of AI capabilities but does not guarantee superior models.