Logprobs
Also called log probabilities, token probabilities
Numbers showing how likely a model rated each token it wrote, and the runners-up, sent back to your code when you ask.
A model picks each token from a ranked list of candidates, each with a probability. Logprobs, short for log probabilities, let your code see those numbers for the tokens it wrote and the ones it nearly chose.
Builders used them to classify with a single token and to flag answers where the model hesitated. Most current reasoning models no longer send them, so those tricks now need another signal.
For example
A support feature asks for a one-word label and reads how much probability the model gave "refund" against "complaint" before routing the email.
Put it to use
Where you'll meet logprobs in ChatGPT, Gemini and Claude, how to use it there, and the mix-up most people make.
Daily Bytes Plus is in the Android app.