If doing a lot of heavy lifting there. Not only is it not a given that they'll get the correct answer for a lot of simpler tasks in fewer tokens, but smaller models are often available at far higher tokens/second inference.

There are certainly tasks where fable will be faster and/or cheaper, but there are plenty of tasks where even Haiku is as fast or faster and cheaper, or where you can e.g. get away with models like gpt-oss that you can get from inference providers providing 10x+ the token/second speed.

If you don't use enough tokens that relying only on Fable becomes a problem, then keep using just Fable. Personally, for my $200/week Max subscription I'd run out of the weekly quota for Fable in a day. At API pricing I'd go bankrupt if I tried doing the things I do with cheaper models using Fable.

same as it ever was. It seems your argument implies a belief that you should always use the best model. Others think that not all tasks require the absolute most powerful, expensive, model.

The CursorBench plot, for example, shows that fable does have slightly better performance, but Opus is pretty close, and is less expensive per task

fable on longer coding tasks with fable subagents will easily chew through hundreds of dollars in a single run.

less expensive per task might also mean less of your own time

People don't seem to be able to reconcile the fact that there is likely an overbuild and overspend on AI that may be inflating a bubble, and that AI is actually incredibly useful and getting really really good for certain tasks. Both camps are right, except for when they say the other is wrong.