The only thing close are the "logprobs": https://cookbook.openai.com/examples/using_logprobs https://cookbook.openai.com/examples/using_logprobs However, comme

The only thing close are the "logprobs": https://cookbook.openai.com/examples/using_logprobs https://cookbook.openai.com/examples/using_logprobs However, commenters around here noted that these have likely not been fine-tuned to correlate with accuracy - for plaintext LLM uses. Would be interested in hearing finding for MLLM use-cases!