alt Hacker NewsSounds pretty sane, the same way how OpenWebUI and probably other software out there also has a concept of “tool models”, something you use for all the lower priority stuff.
Actually curious to hear what others think about why Anthropic is so set on disallowing 3rd party tools on subscriptions.
The sota models are largely undifferentiated from each other in performance right now. And it’s possible open weight models will get “good enough” relatively soonish. This creates a classic case where inference becomes a commodity. Commodities have very low margins. Training puts them in an economic hole where low margins will kill them.
So they have to move up the stack to higher margin business solutions. Which is why they offer subsidized subscription plans in the first place. It’s a marketing cost. But they want those marketing dollars to drive up the stack not commodity inference use cases.