AI researchers at Stanford and the University of Washington developed a competitive "reasoning" model, s1, for under $50 in cloud compute credits, matching the performance of advanced models like OpenAI’s o1 and DeepSeek’s R1.