Fable bumps down to Opus so often for me that I’m pretty much back to just using ChatGPT’s Sol. And Sol will run 13 or 14 hours building without difficulty, too. I think I’m back to being a ChatGPT person, just like that, again.
Some of the bumping is by design. Anthropic’s launch note says that “when Fable’s classifiers detect a request related to cybersecurity, biology and chemistry, or distillation, the response is automatically handled by Claude Opus 4.8 instead.” At launch, TechCrunch reported Anthropic’s early data showed “at least 95% of Fable sessions running entirely on the model’s own responses”. “So often” doesn’t feel like five percent. Whatever the cause, the effect is the same: you ask the frontier model and a lesser one answers, and a builder learns quickly to stop asking.
There’s a money change in the same week. Since Monday, Fable has been off the included weekly limits on Claude plans and runs on usage credits, so any Fable that does answer is now metered. Whether that has anything to do with the bumping, the public notes don’t say.
Sol, meanwhile, is a week old. OpenAI released GPT-5.6 on July 9 and calls Sol its “best coding model yet,” and claims it’s “54% more token efficient when it comes to AI coding tasks.” Back in April 5.5 ran 13.4 hours flat for me without stopping, and Sol keeps that up. A model that answers every time and works all night is hard to argue with. Thirteen hours of unattended building on a flat fee is a different product from a brilliant answer you have to wait for.
There’s no ideology in the switch. Five weeks ago the same logic pointed the other way, when Fable broke a bottleneck that neither Opus nor Codex had cracked in a week.
At this point, you can already pretty much build anything. So I guess now we just look for it to come faster and cheaper.
