Post

running into that "smarter models are actually cheaper because they get the task done faster" problem again. Let my claude subscription run out in favor of an OpenRouter credit limit with various cheaper models.

Watching deepseek v4 flash struggle with formatting a package.json for 30m

Yup, I'm back to a Claude max plan. I long for the day I can self host this sort of thing. Or perhaps my use of frontier models means I'll pretty much always pay for frontier performance

I think the smaller models are fine if you baby them, but the difference between fable and opus is the same difference between opus and these. They spin off onto tangents, don’t deal with ambiguities, and take tasks very literally

My guess is the orchestrator / implementer architecture will work really well here when you can run the expensive/slow model for guiding the cheaper/fast model.