The S1 inference model was developed on the QWEN2.5-32B-Instruct model of China’s e-commerce giant, by researchers from Stanford University where Li works and researchers from Washington University, according to a research paper published last week. I did.
After being trained with 1,000 carefully curated questions and answers to distilled “thinking process” from Google’s Gemini thought experimental model, the S1 model outperformed Openai’s O1-Preview on mathematics and programming skills .
05:00
Does China’s low-cost deepseek arrival mean the end of Nvidia’s chip advantage?
Does China’s low-cost deepseek arrival mean the end of Nvidia’s chip advantage?
The cost of running only a Graphic Processing Unit (GPU) to develop an S1 could be $14 USD based on the calculations described in the study. These tips can be rented for $2 per hour.
