All-you-can-eat, but no signature dish: why Fable 5 left the Max subscription

On July 7th, Anthropic pulled its most expensive model, Fable 5, out of the Max subscription and moved it to pay-as-you-go. The official reason is capacity. This piece is about the bigger thing underneath that capacity: why "the best tool" and "all-you-can-eat" are fundamentally incompatible — and how, at the very moment the frontier gets pulled behind a meter, on the everyday end even your next sentence is starting to be written for you in grey.

A monthly plan is a bet. It bets that your average appetite has a ceiling, and that every bite is cheap enough. A gym can sell memberships because most people won't show up every day; a buffet can open its doors because the cost of the signature dish gets spread across a whole table of cheap sides.

Claude's subscription is the same bet. Pro and Max sell you a month's worth of allowance on the premise that the model you use day to day is cheap enough per token to serve all-you-can-eat. That bet held — right up until Fable 5.

As of July 7th, Anthropic moved Fable 5 out of Max's weekly allowance. To keep using it, you switch to usage credits, metered at API prices. The official reason is capacity: "demand will be very high and hard to predict," with an emphasis on bringing it back into the subscription as soon as compute catches up. That's half the truth.

The other half is clear from the pricing. Fable 5's API price is $10 per million input tokens and $50 per million output — exactly double Opus 4.8, and the most expensive model Anthropic has ever sold. It also "draws down usage faster": the same conversation burns through far more allowance than other models. A dish that costs twice as much, is the most in-demand, and is unusually consuming — put it on the buffet and there are only two endings: the house loses money, or it throttles you until the dish might as well not exist. Switching to metered billing is a way to reconnect something fundamentally incompatible with a monthly plan through "you pay for what you burn."

I have to draw an honest line: capacity is what theystated outright; "too expensive, all-you-can-eat would break the monthly plan" is what I've inferred from the pricing, not something spelled out in black and white. But the two are entirely consistent, and there's a tell — they keep saying they want to bring it back into the subscription. You don't talk that way about a product you've "decided to sell as metered only." It sounds more like: we can't sustain it right now, not that we don't want to give it to you.

What's interesting is that phrase, "hard-to-predict high demand." It's hard to predict precisely because Fable is good enough to make you want to keep using it. All-you-can-eat works because most people actually don't use much; but when a model is so good you can't bear to switch away, the premise that "average appetite has a ceiling" collapses.The best dish is precisely the one that can least afford to go on the buffet. The more people want it, the less possible it is to give it away freely.

This isn't a quirk peculiar to Anthropic; it's the shape of the frontier. You can see it settling into two tiers: one subscription-based, good-enough, predictable everyday brain; and one metered, deepest, expensive frontier brain. You run most of your work on the middle tier, and when you really need that feel, you pay separately, once, for the frontier tier.

And "feel" is something creative people sense before anyone else. Where I most want to keep Fable was never coding or running workflows — Opus was already competent at those. It's when writing: that depth that doesn't rush to slide into the smoothest word, that's willing to pause on the subtle points, that takes a beat longer to close. And that very depth is the most expensive, most consuming, most in-demand thing there is.

So what really ended on July 7th wasn't a promotion. It was the brief, generous illusion that "the best tool can just be bundled into the monthly plan." From here on, the frontier is à la carte. You have to think it through first: which task is worth paying separately, once, for that feel.

One more note, about this article itself.

I wrote it with Fable, in the last few hours before Fable was carried off. And the idea to "write this up as a piece" — half of it wasn't mine. A line of grey text popped up in the input box first, having already thought out my next step, and I glanced at it and hit TAB.

So at that same moment, two things were pulling in opposite directions. On one side, the deepest tool was being pulled behind a meter, forcing you to choose harder just to reach it. On the other, the everyday tool was smart enough to write your next sentence in grey first, shrinking "choice" down to a single key, tempting you to not-choose with less effort.

The frontier getting more expensive is meant to make you seriously decide whether it's worth it. The grey text getting more accurate is meant to make you stop deciding. These two forces don't conflict — they're just asking the same question from both ends: when are you still willing to pay a little more to "think it through yourself, to speak up yourself"?

A monthly plan can't serve a signature dish. And beyond the signature dish, even what you're going to say has already been drafted for you.