Based on what I said. If e.g. Sonnet (assuming it’s significantly smaller than Opus) is unprofitable why are there a bunch of inference providers on OpenRouter serving very large models way cheaper? They don’t have a pile of money to burn for no reason.