Simon Willison's Weblog recently featured a post about OpenAI's new 'Predicted Outputs' feature for GPT-4o and GPT-4o mini. The feature allows users to send a 'prediction' along with their prompt which can accelerate the response time by validating large batches of input in parallel. OpenAI's pricing for this feature will charge for any tokens provided that are not part of the final completion at standard rates. The author shared a result comparison showing a faster response with prediction but at a slightly higher cost.