The Fable Tax Never Left

Anthropic said Fable 5 was included in my Max plan starting July 20th. My terminal spent every session since sending me to a credits page instead. The bug behind it, the pattern behind that bug, and why Opus 4.8 is my default again.

The Fable Tax Never Left — AI

July 20 was supposed to be the boring day. That’s the date Anthropic set for folding Fable 5 back into the Max plan I already pay for — included, the same word that’s always applied to Opus and Sonnet. I opened a new Claude Code session that morning with nothing to write about. Instead I got a screen telling me to set up additional usage credits before I could touch the model I was told was already mine.

I opened another session a few hours later. Same screen. Different session the next day, same screen again. At some point a bug starts behaving like a policy, whether or not it technically is one.

I already did the spreadsheet on this. In The Fable Tax Moved Inside the Plan, I worked out that Fable costs roughly twice what Opus 4.8 costs per token, so “included at 50% of your weekly limit” quietly means the same quota now burns at double speed the moment you touch Fable — the tax didn’t disappear on July 20, it just changed which line item it hides in. That post was about the economics. This one is about what happened when I actually tried to use the thing.

What Anthropic Actually Promised

This part isn’t my read, it’s the text. Anthropic’s own account posted that beginning July 20, Fable 5 would be “included in all Max and Team Premium plans, at 50% of limits,” with Pro and Team Standard staying on usage credits plus a one-time $100 credit. The Claude Help Center article says it plainer: Fable 5 is “a standard part of your plan,” and you can use it “at no extra cost” up to that 50% ceiling. I’m on Max. By that text, July 20 should have been a non-event.

My Experience, and the Part That Isn’t Just Mine

What I saw, across more than one session, was a hard stop: a prompt to set up usage credits before Fable would run at all, on an account with plenty of its weekly allowance untouched. That’s my account, my few days, my recollection — I’m not running a controlled study, and I won’t pretend this is a universal number.

It turns out somebody else’s terminal is on the public record. GitHub issue #79337, filed against anthropics/claude-code the same day inclusion went live, reports a Max account with usage nowhere near exhausted getting told: Switched to Opus 4.8 (1M context) for this session · Fable 5 requires usage credits · /model to change. Running /model opens a browser tab asking the user to top up credits before Fable will run. The reporter’s working theory is an entitlement check that treats the saved alias claude-fable-5[1m] as a literal string instead of resolving it to the base model plus a separate context flag — so the [1m] suffix breaks the match and the system fails toward “pay us” instead of “you’re covered.” The issue links four earlier tickets showing the same seam: the identical symptom during the July 17–19 promo window, two prior bugs specific to that [1m] alias, and a general model-to-billing routing mismatch. That’s not a first-day fluke. That’s a seam that’s been open for weeks.

The community noticed. The July 17 instance of this same bug — two days before the real rollout — produced an 85-comment Ask HN thread before Anthropic patched it. The thread on the July 20 announcement itself mixed relief from people resubscribing with real frustration: Anthropic had promised 14 days of included access back at the June 9 launch, delivered about three before an export-control pause, and came back to a shorter window at half the weekly limit. One commenter put it bluntly: “Not a good look to bring Fable back and then both half the usage and take away days.”

None of this requires Anthropic acting in bad faith. A billing entitlement path that fails closed instead of open is a completely plausible, boring engineering mistake. But intent doesn’t change what it feels like from the paying side: I was told a model was included, and the product kept asking me for money to use it anyway.

A Credits Screen Is Not a Small Bug

A model can be excellent and still be a bad default, because defaults live in the boring part of the day. I open a terminal, hand a task to a model, and keep working — a default earns that position by being available and predictable, not by being clever. The moment a credits prompt shows up, it drags a pile of questions into the middle of actual work: is my allowance really exhausted, is this a bug, will continuing quietly start a separate charge, do I need to top up even though I’m already paying $100 to $200 a month for the plan this was supposed to be part of. I shouldn’t need to debug a billing state to use the thing the announcement told me was already included.

A Familiar Shape

This exact move — promise flat access, then discover reasons it isn’t quite flat — isn’t new to AI dev tooling, and Anthropic isn’t the first vendor to do it this year.

VendorThe PromiseWhat ShippedWhere It Stands
Anthropic — Fable 5 on MaxIncluded at 50% of weekly limits, “at no extra cost” (Jul 20)A public, reproducible bug (GitHub #79337, four related issues behind it) demands usage credits from qualifying Max accountsUnresolved at publish — read as a bug, felt like a broken promise
GitHub — Copilot premium requests → AI CreditsSame premium features, just metered by the token (Jun 1, 2026)The old fallback-to-cheaper-model safety net removed outright; 10–50× cost jumps reported on agentic sessionsNo reversal after the backlash
Cursor — Pro plan usage“500 fast responses, then unlimited slower”Became $20 of metered credits, first described in confusing “rate limit” language; users hit surprise overage chargesPartially fixed — refunds issued, wording corrected

Different root causes, same lesson: a bug that fails toward billing and a policy that deliberately narrows access are indistinguishable from the customer’s side of the screen. “Included in your plan” is now a claim to verify against the actual product, not a line to take on faith from the announcement post.

Why Opus 4.8 Gets the Default Back

My response was deliberately unglamorous: I went into my own configuration and set Opus 4.8 as the default model everywhere, no per-project exceptions. Not because Fable 5 is a worse model — on a chunk of hard benchmarks it probably still edges Opus. Because Opus 4.8 is the one model in this entire saga that has never once stopped me to manage a balance. It’s priced the way I already understand — $5 and $25 per million tokens against Fable’s $10 and $50 — and it already has the long-context option I actually use.

There’s a dry irony in the choice. Fable 5’s own safety classifier already hands a small percentage of sessions back to Opus 4.8 automatically whenever it flags a sensitive request — Anthropic’s own product already leans on this model as its catch net. I’m just cutting out the extra hop and trusting it a step earlier than the product forces me to.

Opus 5 Can Take Its Time

There’s a rumored Opus 5 somewhere in the pipeline — no official announcement, no published model ID, no system card, nothing beyond a research build nicknamed “Honeycomb” that flashed inside a competitor’s model picker for a few hours in early July before it got pulled. If that leak means anything, the guessed window is sometime between now and early August, with a rumored million-token context and a reasoning tier above Fable’s. All of that is speculation dressed as a leak, not a product, and I’m treating it that way until it shows up in Anthropic’s own docs.

I’m in no hurry regardless. Fable spent June getting export-controlled off the map entirely, came back in July, and spent its first week of being “included” still asking to be paid for. Opus 5 gets to launch on its own schedule. I’ll believe the word “included” again once my terminal stops arguing with it.

Hot Takes

“Included” is runtime behavior, not marketing copy. If a paying subscriber selects the advertised model and lands on a top-up page anyway, it isn’t included in the only sense that actually matters to the person typing.

Read the issue tracker before the announcement post. Four related tickets deep on one entitlement bug is a better audit of a vendor’s billing claims than any press release. The receipts are always in the bug reports, not the blog posts.

I’m not owed a frontier model for $100 a month, and I’m not claiming Fable 5 is bad. What I won’t do is spend part of every session negotiating with a billing screen for access I was told already came with the plan. Opus 4.8 doesn’t ask. Right now, that’s the whole pitch, and it’s enough.

📖
Related Reading

The Fable Tax Moved Inside the Plan — the economics of “included at 50% of limits,” written two days before this one.

The Fable Tax — where this series started: why a separate Fable meter broke the subscription.

Advisor and Executor — the pattern for using Fable 5 without eating the Fable 5 bill.

Opus 4.8 Would Rather Tell You It Failed — the model that just got its old job as my default back.
💬
Working with a team that wants to adopt AI-native workflows at scale? I help engineering teams build this capability — workflow design, knowledge architecture, team training, and embedded engineering. → AI-Native Engineering Consulting