NEW OPENAI FEATURE: PREDICTED OUTPUTS

Summary from the TLDR newsletter

The OpenAI API has a new feature called Predicted Outputs. It allows users to send content as a 'prediction' and have GPT-4o or GPT-4o mini use that content to accelerate the returned result. Any tokens provided in predictions that are not part of the final completion will be charged at completion token rates. This means the more an output differs from the prediction, the more it will cost. Using Predicted Outputs can reduce processing time.