Fable 5 Is Back — and the Clock Runs Until July 7
Anthropic's new flagship returned today, and subscribers get one week to drive it on their existing plan. Here's what the model actually is, why it vanished for three weeks, and how I plan to spend my window.
This morning I opened Claude Code, typed two words, and my daily driver changed:
/model fable
Set model to Fable 5 and saved as your default for new sessionsThat's it. No waitlist, no separate app, no new subscription. If you're on a paid Claude plan, the most capable model Anthropic has ever shipped to the public is sitting in the model picker right now — web, desktop, mobile, and Claude Code. The catch is a date: July 7. More on that below, because the fine print is the actual story for anyone paying a monthly bill.
What Fable 5 actually is
Fable 5 is the first model of the Claude 5 family, and it introduces a new tier that sits above Opus — Anthropic calls it Mythos-class. It ships as a pair of twins built on the same underlying model:
- Claude Fable 5 — the generally available one. It carries safety classifiers that can decline requests in a few sensitive domains (research biology, most offensive cybersecurity) and hand the request over to Opus 4.8 instead.
- Claude Mythos 5 — the same model without those classifiers, available only to vetted organizations through a program called Project Glasswing.
The headline specs read like someone maxed the sliders: a 1-million-token context window by default, up to 128K output tokens per request, and reasoning that is always on — you don't toggle "extended thinking" anymore, the model decides how hard to think on its own. It's built for the kind of work where a single request can legitimately run for many minutes: long autonomous coding runs, deep research, tasks that span millions of tokens.
The classifier fallback is genuinely new territory for an API: a declined request isn't an error, it's a normal response with a refusal stop reason — and the API can transparently re-serve it with Opus 4.8. As a subscriber you'll rarely notice; as a developer it's a new branch in your code.
The strangest launch of the year
If the release feels chaotic, that's because it was. Fable 5 has now launched twice:
- June 9Fable 5 and Mythos 5 announced and released. Subscribers get it included in their plans as a launch promo.
- June 12Fable 5 goes offline. An export-control restriction hits Anthropic, and the model disappears from the picker for almost three weeks.
- July 1The restriction is lifted; Fable 5 returns globally. The subscriber window restarts — today.
- July 7The included window closes. From here on, Fable 5 runs on separately purchased usage credits.
I'll admit the June episode annoyed me — I had barely started testing when the model vanished mid-week. Which is exactly why I'm treating this second window differently: not as a curiosity, but as a deadline.
The fine print: 50% of your weekly limit, until July 7
Here's the part that matters if you pay for Claude. Until July 7, Fable 5 is included in Pro, Max, Team, and seat-based Enterprise plans — but not unlimited. You can spend up to half of your plan's weekly usage limit on it:
Two things to know about that gauge. First, Fable 5 drains the meter faster than Opus 4.8 — the same conversation costs more of your weekly allowance, so the 50% goes quicker than you'd think. Second, the cap is also protection: you can't accidentally torch your entire week's allowance on one overnight Fable run. Hit the 50%, and you either switch back to Opus or continue on usage credits.
After July 7, the included allowance ends entirely. Fable 5 stays available to subscribers, but every token runs on purchased usage credits. The API pricing gives you a feel for the tier gap:
Exactly double Opus. Anthropic says it wants to restore Fable 5 as a standard part of subscriptions when capacity allows — but there's no date on that promise, so I'm planning around the one date that exists.
One week, half a tank, the strongest engine available. Don't waste it on small talk.
How I'm spending my week
The advice from people who had early access is consistent: Fable 5 shines on the problems you'd normally break into pieces because no model could hold them whole. So my plan is to feed it the backlog items I've been avoiding, not the daily routine:
- The big refactor in my Godot project that touches half the codebase — the kind of long-horizon change where a 1M-token context stops being a spec-sheet number and starts being the difference between "it remembers the architecture" and "it forgot file three of forty".
- A full-repo review of this very portfolio site — one pass, whole history, every route.
- Routine questions, quick fixes, commit messages? Those stay on Opus. Burning double-rate allowance on "rename this variable" is how you wake up on July 4 with an empty gauge.
Practical tip: check your usage meter more often than usual this week. The combination of "drains faster" and "hard 50% ceiling" means the window can end for you before July 7 does.
Fable 5 is the first model in a genuinely new tier, and for one week it's effectively part of the subscription you already pay for. Whether the usage-credit pricing after July 7 is worth it is a question I can't answer yet — which is precisely what this week is for. Pick your hardest problem, point the strongest model at it, and decide with data instead of marketing. The clock is already running.