OpenAI has successfully reduced the costs of running its AI models by over half, specifically for guest users of ChatGPT. This was achieved through new optimizations that decreased the number of required Nvidia GPUs to just a few hundred. The techniques used to accomplish this are unclear, and it’s uncertain whether these gains can be applied to the full ChatGPT product. This development could lead to improved services, faster responses, or increased profitability. Why it matters: these cost savings could give AI labs the breathing room they need to scale and improve their services without being held back by high operational costs.