Cached Reasoning Safety

A feature introduced in the GPT-6 API that isolates reasoning parameters from cache keys, allowing developers to adjust reasoning behavior without invalidating the prompt cache, thereby balancing flexibility with cost optimization.