Optional reasoning effort for agents that support it. Codex maps this to model_reasoning_effort, Claude Code maps it to --effort, and OpenCode maps it to --variant. Common values include low, medium, high, xhigh, and max. Codex also supports ultra on compatible models, and OpenCode supports minimal. You can instead append a recognized value to the model, such as gpt-5.6-sol:xhigh. This field takes precedence over that shorthand. Unsupported agents ignore this field.