The Churn You Can't Un-Churn: Fable 5 Is Back, the Subscriptions Multiplied Anyway

The fourth blink came Thursday evening, two days before Anthropic’s own deadline: from July 20, Fable 5 is a standing part of Max and Team Premium plans at 50% of limits, while Pro and Team Standard move permanently to usage credits with a one-time $100 credit. I called the capped return when the meter was announced; the bet resolved on its first bullet. The meter died without ever billing a Max subscriber.

This post is about what the settlement can’t settle. A reset restores capacity. It does not restore the thing five weeks of whiplash actually spent, and the receipt for that isn’t a cancellation curve. It’s an invoice stack.

The Defection That Doesn’t Look Like One

Here’s my own: three Claude Max 20x subscriptions, $600 a month to Anthropic. Plus a ChatGPT plan for Sol. Plus a $39 Kimi Code subscription bought this week. Plus a Grok plan. Four vendors, roughly $900 a month, one working professional, all of it arbitraged through a launcher that swaps the model under one harness.

Now look at that stack from each vendor’s dashboard. Anthropic sees a customer who tripled his spend this year. OpenAI sees a retained subscriber. Moonshot sees day-one growth. Grok sees a conversion. Every single retention metric reads me as a win, and every single one is wrong about what happened: no product in the stack can sustain a full working week on its own, so loyalty got replaced by a hedge. The market’s actual churn event was the death of the single-vendor customer, and it’s invisible precisely because it presents as revenue.

Nobody knows if Fable 5 survives Sunday. Nobody knows what the limits are week to week. Nobody knows anything, because they refuse to communicate.

— BridgeMind founder, cancelling $400/month of Claude Max for Kimi

That’s the named version of the same event going the other direction: consolidation instead of accumulation, and the reasoning isn’t price or capability. It’s that planning a work week around a vendor requires knowing what the vendor will sell you next week, and that knowledge stopped being available. A hedge is the rational response to unreliability, and hedges are habits. Nobody un-churns from a hedge; the hedge just quietly rebalances.

Two Exits, One Chart

The metering window produced two kinds of leavers, and only one shows up in anyone’s data. The mass-market response was to download the other app: that’s the step function OpenAI’s product lead posted mid-window. The professional response was to keep the Claude Code harness and swap models underneath it, which registers on no chart anywhere: not as a Claude cancellation, not as a Codex install. The public defection curve therefore understates the leak (harness-keepers are invisible) and overstates it (installs aren’t stayers) at the same time, which means Anthropic made a pricing retreat this week without an accurate measurement of what the pricing cost them. So did everyone else.

The Return Is Also a Sorting

The settlement’s tier split turns the comeback question into two questions:

• Max users have a reason to return the default to Fable. The like-for-like trap cuts in Anthropic’s favor here: anyone who kept the harness kept the muscle memory, and flipping the default back is one config line. This half of the reversal can genuinely work.
• Pro users got a $100 goodbye. For the $20 tier, the ghost town is now zoning: metered still reads as optional, optional still reads as gone. The tier most likely to have downloaded the other app during the window is the tier the settlement writes off.

And both halves very nearly landed on a shrunken floor. Until Friday morning, Anthropic’s own support page had the 50% weekly-limit boost expiring the same morning Fable returns: a one-third cut to the ceiling every subscriber actually lives under, unmentioned in the settlement tweet, spotted by the replies within the hour. Then, one day before the deadline, the fifth blink: the boost extended through August 19 for every paid tier. The cut that was too quiet to announce proved too loud to land.

What Would Prove the Reversal Worked

The falsifiable version, in the arc’s tradition. The settlement succeeds if, by late August:

• Multi-sub stacks consolidate. People like me drop back to one or two subscriptions because one vendor’s ceiling reliably covers a week. I’ll report my own stack honestly either way.
• The Fable default returns. Harness-keepers flip back and stay, because the allowance holds and the limits are communicated before they change, not after.
• Pro churn stays quiet. The written-off tier takes the $100 and doesn’t become next month’s BridgeMind thread.

I’d bet on the first half of the sorting and against the rest. Not because the model isn’t good - Fable remains the best model I’ve used - but because the thing that breaks a hedge isn’t a better week of access. It’s a long enough run of boring, predictable weeks that the hedge stops earning its premium, and predictability is the one product this market hasn’t shipped all year. The subscriptions multiplied under uncertainty. They’ll consolidate under boredom, or not at all.

The Next SKU Is Already Priced

One more prediction, because the arc has earned it. The ratchet this month completed has a known final move: tighten, let the degraded state become the new normal, then sell the old normal back as a premium tier. Max itself was born exactly this way - Pro limits squeezed through early 2025 until a $100/$200 tier appeared to absorb the pain - and OpenAI ran the identical play with its $200 Pro plan. The next rung is due, and multi-sub stacks like mine are the market research: $600 a month for three accounts’ worth of juggling is a standing bid that says what the top of the market pays. A single tier above Max 20x that undercuts the DIY stack consolidates the hedge overnight and books as growth.

So: a tier above Max 20x, priced $400 to $600, marketed at professionals, within a few months. The shape I’d bet on is not a bigger multiplier - a flat 100x gets bought only by people who redline it, and flat-rating Fable to redliners is the exact lesson the meter saga taught. More likely: a priority tier where the limits are contractual and announced in advance (predictability is the scarce good, and it’s free to supply when demand is normal), or a sub-plus-credits hybrid that keeps the meter alive inside the bundle. The tell to watch for is in the marketing: whatever ships will describe June’s ordinary subscriber experience as a premium feature. When that page goes live, this paragraph is either a receipt or a retraction, and I’ll link it here either way.