First I resolved it with retry logic, but I don’t see recently this issue to be popping up. So it seems resolved for me.
Having the same issue. I understand the only way forward is a retry logic and then falling back to the regular completions
Same issue here. I get an error message about the output length limit being reached while it shouldn’t if output format is respected (the expected output is actually very small). Can’t post the prompt for confidentiality reasons, but would be happy to share it with OpenAI’s developers. It fails with gpt-4o but works with o1.