Low-Cost Rival to ChatGPT
The S1 model, developed for just $50, demonstrates how AI systems can be trained at reduced costs. Using distillation, S1 achieves competitive performance, but developing advanced technologies still requires enormous resources.
The emergence of the S1 model, developed by researchers from Stanford and Washington universities, marks an important step in the field of artificial intelligence. This model was trained with a surprisingly low investment of only $50. By using a technique called 'distillation', the researchers were able to teach S1 how more complex models like Google's Gemini 2.0 work. Distillation allows smaller models to learn from larger ones while maintaining similar performance at significantly lower costs. S1 was trained in just 26 minutes and used only 16 Nvidia H100 GPUs, whereas more advanced models require tens of thousands of GPUs.
The S1 model has demonstrated competitive performance in various benchmarks compared to well-known AI systems. Additionally, it utilizes a 'reasoning' approach to improve the quality of generated responses. This innovation has raised concerns among technology giants, especially following the success of DeepSeek, which showed that remarkable results can be achieved on a reduced budget. Reactions from established companies in the sector have been swift, with calls to restrict Chinese entities' access to U.S. technological resources.
Despite the advancements, researchers warn that developing truly advanced artificial intelligence models requires significant resources. While distillation allows for the replication of existing model capabilities at lower costs, creating new technologies still necessitates investments that only industry leaders can afford. The United States is investing $500 billion in the Stargate project to maintain its leadership position in AI, highlighting the importance of adequate resources for future development. In this new context, anyone can approach the creation of advanced language models with a limited budget, but significant challenges remain for innovation in the sector.
