Defaulting every ticket to the flagship model is how bills explode. Routing easy tasks to cheap models is how you stop paying the frontier tax. The tax is sometimes worth it: hard coding, long-horizon agents, or evals you cannot afford to fail.
The extra cost — tokens, latency, or dollars — of using a frontier model when a smaller one would have been enough.
Defaulting every ticket to the flagship model is how bills explode. Routing easy tasks to cheap models is how you stop paying the frontier tax. The tax is sometimes worth it: hard coding, long-horizon agents, or evals you cannot afford to fail.