cost control is a policy problem - we certainly don't need to use opus 4.6 for a simple test refactor, but many people (including myself) default to it anyways.

cost control is a policy problem - we certainly don't need to use opus 4.6 for a simple test refactor, but many people (including myself) default to it anyways. we need a way to measure cost / performance for agents on individual repos, with individual types of tasks, to get a better sense of what tasks can be trusted to cheaper agents, and what tasks must be routed to the SOTA