ModelStopReason

Purpose and behavior

Why a model result ended, normalized by each model provider in ModelResult.stopReason. Values: “end” (final answer), “tool-calls” (the model requested tool calls), “max-tokens” (output or context window limit reached; the built-in harness fails with code limit), “refusal” (refusal or content filter; the built-in harness fails with code response).

Complete example and detailed rules.

Signature

export type ModelStopReason = "end" | "tool-calls" | "max-tokens" | "refusal";