ModelReasoning

import type { ModelReasoning } from "@elie-laloum/outpost";

Purpose and behavior

Reasoning effort of an AgentModel or ModelRequest. Values, from least to most effort: “none”, “minimal”, “low”, “medium”, “high”, “xhigh”, “max”. The executing harness or model provider rejects an unsupported level when the agent is composed: Claude Code and Codex accept “low” to “max”, the Anthropic provider every value except “minimal”, the OpenAI provider forwards the value as is, and the other CLI agents accept none.

Complete example and detailed rules.

Signature

export type ModelReasoning =
  "none" | "minimal" | "low" | "medium" | "high" | "xhigh" | "max";