Classification

classifier.dev assigns caller-supplied labels to text. The same input can be classified once, against several independent labels, or across named dimensions.

Language

Input: One text supplied for classification. Its position identifies its corresponding result.

Label: One category the caller supplies. Single-label classification chooses one; multi-label classification can choose several or none.

Dimension: A named classification task with its own labels and optional instructions. Each input receives one decision for every requested dimension.

Decision: One classification of an input against a label set. An input with three dimensions requires three decisions.

Scores: The model's probabilities for the supplied labels. Single-label scores form a distribution; multi-label scores describe independent choices. They do not establish whether the supplied labels cover the input.

Confidence: The model's confidence in its selected label. It can differ from that label's score, and is unavailable when no comparable estimate exists.

Escalation: Re-asking an uncertain decision with a reasoning model. The new answer does not inherit the first model's confidence or scores.

Fallback: Using another model because the preferred model could not answer. This differs from escalation, which starts with a valid but uncertain answer.