Researchers from Stanford University and the University of Washington have unveiled an AI reasoning model named “s1,” trained at a cost of less than $50 in cloud computing credits.
This achievement, detailed in a recent research paper, demonstrates that high-performance AI models can be developed with minimal financial resources (via TechCrunch).
The s1 model exhibits performance comparable to leading reasoning models, such as OpenAI’s o1 and DeepSeek’s R1, particularly in mathematics and coding assessments. The researchers have made s1, along with its training data and code, publicly accessible on GitHub, promoting transparency and collaboration within the AI community.
To develop s1, the team employed a technique known as “distillation.” This process involves fine-tuning a base model by training it on the outputs of a more advanced AI model, effectively transferring its reasoning capabilities. In this instance, s1 was distilled from Google’s Gemini 2.0 Flash Thinking Experimental model.
he training process for s1 was remarkably efficient. The researchers curated a dataset of 1,000 carefully selected questions, each accompanied by detailed answers and the corresponding reasoning processes derived from Gemini 2.0. Training s1 took less than 30 minutes using 16 Nvidia H100 GPUs.
Niklas Muennighoff, a Stanford researcher involved in the project, noted that renting the necessary computing power would cost approximately $20 today.
A notable innovation in s1’s design is the incorporation of “test-time scaling,” which allows the model to allocate more time to think before responding to a question. By instructing the model to “wait” during its reasoning process, the researchers enabled s1 to produce more accurate answers.
The success of s1 raises important questions about the accessibility and democratization of AI technology. The fact that a small team of researchers, with limited funding, can develop a model rivaling those produced by industry giants challenges the notion that cutting-edge AI development is exclusive to well-funded organizations.
This advancement also brings to light concerns regarding the commoditization of AI models. If high-performance models can be replicated with minimal investment, the competitive advantage held by leading AI labs may diminish.
Want to see more of our stories on Google?
P.S. Want to keep this site truly independent? Support us by buying us a beer, treating us to a coffee, or shopping through Amazon here. Links in this post are affiliate links, so we earn a tiny commission at no charge to you. Thanks for supporting independent Canadian media!
