You aren't paying API prices. Anthropic (or OpenAI, SpaceX, etc) are almost certainly losing money on your subscription. How long they'll be able to continue to subsidize your usage is an open question, but they definitely won't be able to do it with everyone (currently, their solution is to make enterprise users pay per token, but that might change). What that means is that for the customers that the frontier labs need to be profitable, going with a closed weight provider isn't "marginally better intelligence for marginally higher (or even lower) prices" but "marginally better intelligence for much higher prices".
This.
I get (sometimes) frustrated with people that don't work in the tech industry over this concept, and having them rationalize that their usage is heavily subsidized by the distributor of the model.
The costs in compute, electricity, cooling, etc. per token is much higher than most people (I feel) who work in the industry realize. Which, cognitively makes sense: they just want the 'magic 8-ball' to give them a tasty dinner recipe that meets their specifications, for example, and don't care how the 'sausage' is made.
However, economies generally equalize with supply and demand, and this (most likely) will become cheaper as the technology progresses and optimization passes/new algorithms are discovered.
If you look at what it costs to serve Kimi K3, the exact opposite of what you claimed is the case.
It seems likely that Anthropic and OpenAI's inference margins on API prices are >95%.
That also casts doubt on the parent's claim that subscriptions are subsidized. It rather seems almost certain that they are overall generating profit.
Exactly, like it has done in cycle after cycle in the past. It would be very naive to assume that once you have a moat it will stay that way with a thing like this.
Unless new data has come out, we don’t know if they are losing money on subscriptions. We know that there is a big difference between subscription and API costs per token, but that alone doesn’t mean much. Subscriptions may be underutilized, or margins on API tokens may be very high. Both are probably true.