OpenAI randomly gives the score -10.82491 when classifying text

Dear community,

I’ve created a classification fine-tune which used to work quite smoothly. Now it seems like randomly classifications are given either the score -10.82491 or a correct one.

Do you know why this is the case? Is OpenAI making changes to their setup like updating their model or servers? Or can it not handle many requests in a row - does it need a cooldown period between each request?

Best regards,
Frederik