model-hierarchy

Model routing reduces AI agent costs by classifying tasks by complexity and directing them to the cheapest capable model. It applies tiered classification across agent workflows to minimize spend without sacrificing results.

To set up tiered model routing, copy the SKILL.md file to your skills directory and restart the gateway to load the skill. This enables cost-aware model selection and dynamic spawning of sub-agents based on configurable task signals.

Yes, you can route tasks to different models in OpenClaw workflows. The skill classifies tasks into routine, moderate, and complex tiers, applying cost-aware decision making to select the cheapest capable model for each specific task.

Vision tasks are handled through configurable vision overrides in the tiered classification algorithm. This ensures that specific task signals bypass standard complexity routing to use models capable of processing vision inputs.

When a cheaper model cannot handle a routed task, the skill includes escalation handling to manage failures. It uses configurable task signals to dynamically escalate the task to a more capable tier model.

Yes, the skill includes testing workflows and cost calculations. It provides examples with OpenClaw integration and guidance for Claude and Codex to verify that the tiered routing algorithm satisfies functional requirements.