I've been testing Fable 5 and wondering why exactly it feels better than Opus 4.8 and it's hard to pinpoint it, but it is undeniably better. I've seen other people mention that it is more thorough/autistic but in a kind of exploratory way, which aligns with what I've found in my experiments. It seems like a really good model for high-level infrastructure work, the kind of first time work that I mentioned in this message. The meta seems to be use Fable for early high-level architectural/directional decisions, then change to cheaper models for work that is repeated. In my case, creating a new ability type or modifier type should use Fable, creating abilities or modifiers that simply extend existing ones should use Opus. In the average roguelite like Isaac, about 30% of the items are "new gameplay" while the other 70% are derivative. Fable would be good for generating the infrastructure for the 30%, which makes the 70% cheap for another model to implement correctly. This gives you a rough estimate of spend over the lifetime of such a project, at least when it comes to items/passives/etc.

It is a very expensive model, though. I've been on the 5X $100 a month plan since the start of the year and I have to work pretty hard to run out of tokens on Opus, given that the nature of my work is mostly back and forth with the model and I rarely let it run loose. But with Fable this back and forth runs out fast. I'm tempted to buy the 20X plan but I can't justify it with Fable going away in 4 days, and I certainly can't justify using it through the API. I think what I'll do in the next 4 days is just use Fable to set direction for a bunch of different things I want to try and generate thorough code for everything where it can and then continue that work with Opus until Fable is available on the subscription again. But yea, it's definitely a smarter model and I think I also like its personality better, but I haven't tested that that much yet.