Maximizes the model’s reasoning depth for complex or ambiguous problems. Expect higher latency and token usage. Serialized as "high".
"high"