The maximum total number of output tokens the agent can generate across all model calls within a single invocation.