The number of new model releases just now is pretty good, but I find it rather tiresome to keep trying them out, which I have scarcely done since year start and kept mostly to Claude. If Fable does indeed go away and I'm left with Opus 4.8, I'll probably check out the new Elon model, I've read it's pretty fast and I could use a fast alternative since Claude tends to be the slowest of them all. I have no interest in Sam Altman's products, and the Chinese ones are a good fallback if I run out of Opus 4.8 tokens. It's important to not do work for these labs, any more than you're already doing by giving them your codebase, so building harnesses, benchmarks, testing models against one another, etc, is pointless unless it serves your direct goals, in my opinion. Back when actual programming mattered you could get lost being an architecture astronaut, the guy with the engine who has not shipped a game and so on, the same failure mode exists for AI by focusing too much on the details of these tools instead of actually trying to do something with them. Usually this is a sign that the individual lacks vision and doesn't know what to build, which is reasonable, as even I, the greatest creative mind present on this website, sometimes have trouble coming up with ideas for what to do with them beyond my rather small (relative to what could be done) goals of making indie games.