There was a comfortable rule these past few years: don’t try too hard to optimize your setup. The next model is coming soon, it’ll cost the same or less, and it’ll just paper over your problems. Developer Drew Breunig calls that the free lunch. And he’s convinced it’s over, thanks to Fable.
His point: Fable 5 is a great model, no argument there. But it’s so expensive that the old comfort no longer pays off. For most code, Opus is good enough – and so are other models like GPT-5.6, K3 or GLM. So his team started asking a question nobody had needed before: which work belongs on which model?
The numbers line up
That same picture shows up in an analysis people talked about a lot this week. The Ramp AI index estimates, from company credit card data, what money actually gets spent on. For Anthropic in July 2026, the leader is Opus 4.8 at around 28 percent. The pricey top models, Fable 5 and Opus 5, sit at 8 and 3.5 percent.
Put differently: people pay for Anthropic’s strongest models – but rarely use them. They reach for the cheaper one that’s enough for the job. That’s not a knock on Fable, it’s just economics.
What it means for you
I think this idea matters because it shifts how you work. The best strategy used to be: wait. Now the best strategy is: sort. Not every task needs the most expensive model. The heavy reasoning problem maybe does; the rest often doesn’t.
In practice: look at which of your work truly needs top-tier performance and which comes out just as well on a cheaper model. That’s where the leverage is right now – not in waiting for the next announcement, but in smartly routing what’s already here. Which, by the way, fits what ARC-AGI-3 showed this week: often it’s not the model that decides, but what you build around it.
Sources: