I would like to request the addition of a seed-setting feature to the GPT API.
Currently, the GPT API allows for some control over randomness using parameters like temperature and top_p. However, even with identical inputs and settings, the output can vary slightly with each call.
If we could explicitly fix a seed value—similar to how it’s done in image generation models like DALLEit would greatly improve the reproducibility of results, making debugging and quality testing much more efficient.
This feature would be especially valuable in use cases where consistent output is critical, such as automated testing or response verification.
I hope the GPT API will support seed control in a future update.