loading
loading
A single dial that controls how hard the model works — how much it thinks, how many tool calls it makes, how verbose it is.
Set via output_config: {effort: "low" | "medium" | "high" | "xhigh" | "max"}, effort trades intelligence against latency and token cost in one parameter. Lower effort means fewer, more-consolidated tool calls, less preamble, and terser output; higher effort buys deeper reasoning and better self-verification. high is the usual sweet spot and xhigh is the default in Claude Code for coding/agentic work; reserve max for cases where correctness matters more than cost. The relationship isn't always monotonic — on agentic tasks, higher effort up front often reduces total turn count and cost.
Plainly
Think of Effort as the brain part that guesses or decides. A single dial that controls how hard the model works — how much it thinks, how many tool calls it makes, how verbose it is.
In practice
Use it when model choice, prompts, latency, cost, or quality affect the product result. In practice, define the owner, input, output, and failure mode before you rely on it.