[OpenAI Finds Optimization Solution to Halve Inference Costs] According to DeepTech TechFlow on June 30, The Information reported that earlier this month, an insider revealed that OpenAI engineers informed some colleagues that they had developed a set of new optimization techniques capable of reducing model inference costs by more than half. After applying these new techniques to scenarios where visitors use ChatGPT without free or paid accounts, the engineers were able to temporarily reduce the required number of NVIDIA GPUs to just a few hundred.