[tempaccount420]

> They do inference at a loss

They don't, inference is cheap, especially for agents because of cache hits. The API prices are just inflated.