I am making batch requests to v1/chat/completions gpt-4o-mini using a custom json schema. Once response is finalized the website shows “27326 completed, 0 failed of 27326 total requests”. !BUT that is incorrect: the output file contains 22k responses and error file contains the remaining 5k with: “error”: {“message”: “The server had an error while processing your request. Sorry about that!”, “type”: “server_error”, …}.
I observed this behaviour consistently on every batch submission (over 20 submissions) in past few days.