The Fable Tax Moved Inside the Plan

Two weeks ago I argued that Anthropic's plan to pull Claude Fable 5 out of subscriptions and bill it separately wouldn't hold. It didn't. But the way they walked it back is more interesting than the reversal itself — because the tax didn't actually go away. It moved.

The receipt

In The Fable Tax, I made one specific prediction:

Effective July 20, that's exactly what happened. Called it. The interesting part isn't that Anthropic reversed course, though — it's how it reconciled a model priced at twice the rate of Opus with plans that already promise you a usable allocation. The whole trick hides in four words: 50% of limits.

What Anthropic actually announced

Max and Team Premium subscribers get Fable 5 as part of the plan, available at 50% of limits. Pro and Team Standard get access through usage credits plus a one-time $100 credit — a softer landing, not the same inclusion. For Max, the change is simple: Fable no longer needs a separate metered bill stapled to your subscription. It's back inside the product you thought you were buying.

Anthropic frames it as capacity — “demand for Fable has been challenging to predict” — which explains the staged rollout but not the 50% figure. Pricing explains that.

“50% of limits” is the price, reconciled

Fable 5 lists at $10 per million input tokens and $50 per million output. Opus 4.8, already included, is $5 and $25. Fable is exactly twice the price of Opus on both sides — and that symmetry is the whole design.

Anthropic isn't eating the premium. It's translating Fable's 2× price into the plan's own consumption model. Picture your shared weekly allowance as a budget every model spends from, just not at the same rate: an Opus token costs one unit, a Fable token costs two. An all-Opus workload runs until the budget is gone; an all-Fable workload hits the same economic wall at half the token count. No absolute allowance has been published, so any figure in hours or prompts would be fiction. The number that matters is the ratio — Fable gives you half the nominal runway because it burns the shared budget twice as fast.

That's why “included” is honest and “free” isn't. The Fable tax didn't disappear; it changed form — from a separate invoice after you've already paid for Max, into a 2× draw against the subscription you already have. Still a real upgrade: the cost is bounded by the plan again, and you can use Fable without opening a second meter and wondering whether this month's AI bill has developed ambitions.

The meter never really stuck

The metering policy was supposed to establish Fable as a separately paid premium. In practice, Anthropic kept suspending its own rule. Fable stayed included up to 50% of weekly limits as a “promotion” — extended from July 7 to July 12, then to July 19, and now made permanent on July 20. Three extensions followed by permanence isn't a transition to metering. It's a meter being repeatedly asked to wait outside.

The capacity story fits: demand was hard to predict, capacity arrived in stages, access widened as it did. The temporary policy reads less like pricing conviction and more like operational insurance during an uncertain launch. Competitive pressure did the rest — OpenAI's Sol lists at $5 and $30, undercutting Fable's $10 and $50. A surcharge is easy to defend when customers have nowhere comparable to go; a real competitor changes that math fast. People tolerate capacity limits. They're less patient about paying a membership and then finding the interesting model behind a second cash register.

What a Max subscriber actually gets

Max 5× is $100 a month for five times Pro usage; Max 20× is $200 for twenty times. Max carries two weekly limits — one shared across all models, one reserved for Sonnet. Fable draws from the shared all-model pool; the Sonnet-only bucket adds no Fable headroom. Net, a Max subscriber now gets three things: Fable inside the existing plan, no separate Fable bill, and a Fable ceiling that reflects its 2× economics.

Lean on Fable and you'll burn the shared allowance twice as fast as Opus-equivalent work; mix models and your runway depends on the mix. There's no honest universal answer in hours or tasks — workloads differ and the absolute limits aren't public. The precise version is just this: whatever weekly model budget your Max tier gives you, Fable has half the token runway of a model priced at the Opus rate.

Why the tax was never going to hold

The problem with separate Fable metering was never that Fable is expensive. Expensive models are expensive — compute has resisted motivational speaking. The problem was that it broke the one thing a subscription actually sells: a predictable ceiling. Nobody paying $100 or $200 a month necessarily expects unlimited access to every model. They expect the plan to convert uncertain usage into a bounded number. Anthropic can make Fable eat that budget faster. What it can't comfortably do is sell a premium subscription and then treat its premium model as an unrelated utility bill.

The hot take: a premium AI subscription that excludes the premium model people actually want isn't a premium subscription. It's a cover charge for the meter. Customer resistance exposed the mismatch, capacity made the reversal possible, and Sol made stalling expensive.

The value proposition won

The original call held, with one honest refinement. Anthropic didn't erase Fable's higher cost — it embedded it into the Max plan as a clean 2× consumption rate with a 50% ceiling. The economics survive; the second bill doesn't. That's the version I'd have signed off on two weeks ago, and it's the version subscribers should want: not the interesting model quarantined behind its own meter, but priced honestly inside the plan you already pay for. The tax didn't get repealed. It got absorbed — which, for anyone who refused to run a second meter, is close enough to winning.