OpenAI cuts AI running costs by over 50% with system tweaks

OpenAI engineers say they've cut the cost of running AI models by more than 50% through a series of system-level improvements. The cost refers to the computing power needed when the model responds to user requests.

The improvements mainly come from making better use of existing server resources, not adding new chips. This means fewer Nvidia chips are needed to run their AI systems, and the savings could be used to lower API prices or increase user limits.