ReferenceAsk

Decision

Models that answer typed questions about content as calibrated probabilities.

The decision models behind evaluateDecision. Instead of writing an answer, a decision model reads a state (text, JSON, or for some models images) and answers questions you define: yes or no, one choice from a set, or a score on an ordered scale. Each answer comes back as a probability, so an app can set its own threshold instead of parsing prose. Billing is on input tokens only. Probabilities aren't comparable across models or versions, so tune any threshold against the exact model it will run on.

6 models