I also use them per-token (and strongly prefer that due to a lack of lock-in).
However, from a game theory perspective, when there's a subscription, the model makers are incentivized to maximize problem solving in the minimum amount of tokens. With per-token pricing, the incentive is to maximize problem solving while increasing token usage.
I don't think this is quite right because it's the same model underneath. This problem can manifest more through the tooling on top, but still largely hard to separate without people catching you.
I do agree that Big Ai has misaligned incentives with users, generally speaking. This is why I per-token with a custom agent stack.
I suspect the game theoretic aspects come into play more with the quantizing. I have not (anecdotally) experienced this in my API based, per-token usage. I.e. I'm getting what I pay for.
I saw a funny skit where if free Claude instance was down for you, you could just ask Rufus, Amazon's shopping AI assistant, your math/coding question phrased as a question about a product, and it would just answer lol.
In my region a certain small bank had an AI assistant which someone neglected to limit, so you could put whatever there and not even phrase it as a question about a product.
I personally prefer per-token, it makes you more thoughtful about your setup and usage, instead of spray and pray.
You can also access the notable open weight models with VertexAI, only need to change the model id string.