Log Probabilities (Logprobs) measure how confident a language model is when predicting each token in a generated response. Higher log probabilities generally indicate greater confidence in the selected output.
Voice AI developers use Logprobs to evaluate model confidence, compare alternative responses, improve prompt engineering, and support AI evaluation in production systems.