Fable 5 Is in the Claude Subscription for Good — Max Terms, Limits, and What Pro Users Should Do

The 30-second version

On July 18 Anthropic announced its final decision, and it's already live: as of July 20, Claude Fable 5 is included in the Max ($100 and $200) and Team Premium ($125) plans on a permanent basis. The model can consume up to 50% of the total weekly limit on those plans.

If you're on Pro ($20) or Team Standard ($25), there's no bundled access. Instead, Anthropic hands out a one-time $100 in usage credits — a prepaid balance that token charges draw down from — after which Fable 5 bills at API rates: $10 per million input tokens and $50 per million output tokens.

Nothing changed for the API or for consumption-based Enterprise. The headline: no more "until July 7, 12, or 19." Choosing between Fable 5, GPT-5.6, and Opus 4.8? See which one fits your budget and workload.

What happened: the end of the deadline shuffle

Fable 5 launched on June 9, 2026 as the first Mythos-class model Anthropic deemed safe enough for broad access. Three days later, on June 12, Anthropic pulled Fable 5 and Mythos 5 over US export-control requirements.

Access came back on July 1 — first "until July 7," then extended to the 12th, then the 19th. Every extension spawned a fresh wave of headlines asking the same question: what happens next? So yes, Fable 5 is back — and this time for real: as of July 20 access runs on permanent terms, not inside yet another temporary window.

On July 18 Anthropic's official X account posted the final answer:

The key phrase is will be included — included, not extended. This isn't another temporary window. Since July 20 it's the working reality, not just an announcement: the rollout happened on schedule.

Why did Anthropic jump from "we'll see how capacity goes" straight to "permanent"? Rozetked and a few other outlets suggest that pressure from GPT-5.6 Sol forced Anthropic's hand. That's media speculation, not an official explanation — but it tracks: a competitor shipped a strong model, and Anthropic couldn't afford to leave its flagship on pay-as-you-go only.

How long is Fable 5 available? Indefinitely, now

For Max and Team Premium owners the answer is short: there are no more deadlines — and July 20 confirmed it in practice; access is running normally. Every article that says "Fable 5 is available until July 19" went stale on July 18.

The reason to believe "permanent" is Anthropic's own wording. The company changed the terms four times in a month: shutdown, return, extensions, finale. After that kind of whiplash, "will be included" reads differently from yet another "extended." Some caution is still warranted: if capacity runs dry again, the rules could be revisited. For now the official position is permanent inclusion — and the first checkpoint, July 20 itself, passed with no revisions.

Terms for every plan: one table instead of ten news posts

Prices verified against the official pricing page on July 21 — unchanged since this article was published. If you're picking a plan now, go by your workload, not the model's name.

Pro fits if you use Claude Code a couple of hours a day, not every day, and don't mind paying for Fable 5 separately. The $100 in credits will let you try the flagship without upgrading.

Max 5x is the workhorse choice for a solo developer who lives in Claude Code daily. For $100 a month you get Fable 5 bundled at half the limit, plus five times the headroom of Pro.

Max 20x only makes sense under genuinely heavy load: long agentic sessions, big refactors, parallel tasks. For most people, $200 is overpaying.

Team Premium pays off for teams that want the admin console, Claude Cowork, and Max-level limits per member. Team Standard stays on regular limits with Fable billed per use.

The $100 credits for Pro: how far they actually go

$100 in credits sounds generous, but at Fable 5's prices it burns fast. Let's do the honest math.

At API rates:

• $10 per 1M input tokens;
• $50 per 1M output tokens.

The pure arithmetic: $100 = 4M input tokens + 1.2M output. Or, say, 2M input + 1.6M output.

An agentic Claude Code session with a fat context looks different. Three realistic scenarios:

1. An average session: 200K input + 50K output = $2 + $2.50 = $4.50. That's roughly 22 sessions per $100.
2. A heavy session over a large codebase: 1M input + 200K output = $10 + $10 = $20. That's 5 sessions per $100.
3. With prompt caching: if most input tokens are cache reads at $1/M, then 8M cached input + 1.6M output = $8 + $80 = $88. Nearly the whole credit.

Bottom line: $100 covers a serious test drive or a handful of hard tasks. A full month of daily work — no. If Fable 5 becomes your main model, you'll either move to Max or keep topping up usage credits.

The 50% weekly cap: what it means in practice

Even on Max and Team Premium, Fable 5 isn't unlimited. Anthropic capped it at 50% of the subscription's total weekly limit.

What that means: Fable 5 draws from the same pool as every other model, but can consume no more than half of it. If your weekly limit is X tokens, Fable gets at most 0.5X.

There's a fresh detail here that's easy to miss. On July 20 Anthropic didn't just make Fable 5 permanent — it also ended the 50% weekly-limit boost Claude Code had been running: that same day, limits snapped back to standard. Per The Decoder, for Max and Team Premium that's a cut of roughly a third from the boosted volumes. The practical consequence: the half-limit for Fable 5 is calculated from the already-shrunken pool, so your real flagship headroom is more modest than the first weeks of July suggested.

Why 50%, specifically? Because Fable chews through limits noticeably faster than the regular models. Several independent reviews — t-j, diffnotes, rozetked — note that long Fable sessions hit the quota markedly sooner than the same work on Sonnet or Opus.

Practical advice: don't try to keep Claude Code on Fable 5 full-time. Fire up Fable for the hard stuff — architecture decisions, big refactors, multi-file bugs. Leave the routine work, boilerplate generation, small edits, and documentation questions to Sonnet 5 or Opus 4.8. That way the 50% goes toward what Fable is actually for.

Fable 5 in Claude Code: who actually needs it

Claude Fable 5 isn't just "the next Anthropic model." It's the first public Mythos-class model. Per Anthropic's numbers, Fable 5 scores 80.3% on the SWE-bench Pro benchmark versus 69.2% for Opus 4.8. An 11-point gap on professional real-world coding tasks is a serious jump. For how Fable 5 stacks up against GPT-5.6 and Opus 4.8 across the full benchmark suite and in real work, see our comparison, Fable 5 vs. GPT-5.6 and Opus 4.8.

In Claude Code, you switch to Fable 5 with the /model fable command — you'll need Claude Code 2.1.170 or newer (as of July 21 npm has 2.1.216, so after a claude update you clear the bar with room to spare; on an older version the model simply won't show up in the list). The model can sustain long autonomous sessions, checks its own work, and handles context up to 1M tokens on paid plans.

But here's the honest question: do you need Fable 5?

If you're just getting started with Claude Code, you most likely won't feel the difference between Fable 5 and Opus 4.8. Not because it's small, but because in the first weeks the model matters less than how you brief it. A concrete prompt, the right context, clear constraints, breaking work into steps — that's what saves hours.

So start with Pro at $20. It comes with Claude Code, Opus 4.8, and Sonnet 5. Once you're regularly hitting the limits, or you notice Fable solving problems Opus can't, that's when Max is worth a look.

You'll feel the difference between Fable 5 and Opus 4.8 later. What you'll feel first is the difference between knowing how to brief Claude Code and not knowing — and that skill works exactly the same on a regular $20 Pro plan, no $200 Max required.

Buying a Fable 5 subscription when you're not in the US

A note of context: we build skillmake from Russia, where Anthropic doesn't operate — Russian cards aren't accepted and local IPs are geoblocked — so subscribing means workarounds: a foreign bank card plus a VPN, or reseller services with a 20–30% markup. In ruble terms, at roughly 78 RUB per dollar plus card fees, Max 5x lands around 8,000–9,000 RUB (~$100–115) a month and Pro around 1,700–2,000 RUB (~$22–26); topping up Fable 5 tokens beyond the plan goes through the Anthropic Console balance with the same foreign card. We covered the full mechanics in separate (Russian-language) guides on using Claude from Russia and getting a card that works abroad. If you're reading from a supported country, none of this applies — just subscribe on claude.com.

Fable 5 vs. Mythos 5: what's the difference

Technically, they're the same model. The only difference is the protective classifiers.

Fable 5 is the version with safety classifiers enabled for everyone. When a request lands in a sensitive category (cybersecurity, biology, chemistry, distillation attempts), the answer comes not from Fable 5 but from Claude Opus 4.8. Anthropic says the fallback triggers in fewer than 5% of sessions on average.

Mythos 5 is the same model without those restrictions. It's available only to approved organizations under Project Glasswing, often in collaboration with the US government. A regular user won't get Mythos 5.

The classifiers are the source of the "muzzle" complaints: in rare cases Fable hands the answer off to Opus instead of replying itself. Anthropic says false positives will decrease over time.

Fable 5 unavailable: decoding "is currently unavailable" and other errors

If Fable 5 has vanished from your model list or throws an error, it's almost never an outage — it's one of the by-design restrictions. Let's walk through the common cases.

"Claude Fable 5 is currently unavailable." Four real causes:

• The model isn't part of your plan. On Pro and Team Standard, Fable simply isn't served without usage credits enabled — go to Settings → Billing and turn credits on. After July 20 this is expected behavior, not a bug.
• An outdated Claude Code. Versions older than 2.1.170 can't see the model — update with claude update (2.1.216 as of July 21) and pick /model fable again.
• A load spike on Anthropic's side. Before July 20 the model really was pulled when capacity ran short; these days such episodes are isolated — check status.anthropic.com and retry in a few minutes.
• You've exhausted the 50% weekly cap on Max or Team Premium. Until the week resets, Fable is closed while the other models keep working.

"Fable 5's safeguards flagged this message." Not a ban, not a malfunction. The request tripped the protective classifiers — the same ones from the Mythos section: the reply will come from Opus 4.8, or you'll get a warning. The fix: rephrase without the trigger wording, or split the task into parts. Your account is fine; this fires in fewer than 5% of sessions.

"Did they shut off Fable 5 again?" No. As of July 20 the terms are permanent and in effect right now. If the model disappeared from your list, it's one of the four causes above — not a new shutdown.

FAQ

Will Fable 5 get pulled again?

Officially, Anthropic says that as of July 20 Fable 5 is included in Max and Team Premium on a permanent basis. After a month of deadline whiplash that sounds convincing, though the company reserves the right to change terms. So far everything has gone as announced: on July 20 the model switched on in the plans as promised. Watch the Anthropic blog and the @claudeai account for updates.

Is Fable 5 free — for real?

No. There is no free plan for Fable 5. The "free" in the headlines refers to a one-time $100 in usage credits for Pro and Team Standard users. Once those run out, Fable 5 is billed at API rates.

What happens on Pro after the $100 runs out?

After the $100 credit is spent, Fable 5 on Pro keeps working only with usage credits enabled. Every million input tokens costs $10, every million output tokens $50. With usage credits switched off, Fable 5 becomes unavailable while the rest of the subscription's models keep working within their usual limits.

Why am I seeing "Fable 5 is currently unavailable"?

Most of the time it isn't an outage. On Pro and Team Standard the model isn't served without usage credits enabled; an old Claude Code (before 2.1.170) can't see it; or you've burned through the full 50% weekly cap on Max or Team Premium. A full breakdown of every cause, with fixes, is in the "Fable 5 unavailable" section above.

How much does Fable 5 cost in rubles (for readers in Russia)?

There's no separate price tag for the model — you pay for the subscription. We build skillmake from Russia, so for us the local math matters: at roughly 78 RUB per dollar plus payment fees, Max 5x ($100) works out to around 8,000–9,000 RUB (~$100–115) a month, Pro ($20) to about 1,700–2,000 RUB (~$22–26); beyond that, Fable 5 tokens run $10 per million input and $50 per million output. Resellers charge a markup on top. More in the section on buying a subscription from Russia.

What does "Fable 5's safeguards flagged this message" mean?

Your request tripped Fable 5's safety classifiers — the reply will come from Opus 4.8, or you'll get a warning. It is not an account block or a ban: rephrase the request or split the task into parts. It fires in fewer than 5% of sessions.

Fable 5 or Opus 4.8 — which should a beginner pick?

A beginner will do fine with Opus 4.8. It's cheaper, faster on simple tasks, and doesn't require Max. Fable 5 shines on complex, long, multi-file work. While you're still learning to write prompts and finding your way around Claude Code, the gap between models will matter less than the gap between a good request and a bad one. A detailed model comparison is in our separate breakdown.

Can I download Fable 5 and run it locally?

No. Fable 5's weights are closed; the model can't be run locally. If you want a strong open-weights alternative for local use, look at Kimi K3 — we covered it separately.

Does Fable 5 work in Cursor?

Yes, but only via the API. In Cursor you can configure the claude-fable-5 model with your own Anthropic API key and pay for tokens as you go. A Claude.ai subscription can't be connected to Cursor. More on choosing between Claude Code and Cursor — in our "Claude Code or Cursor" comparison.

This article is part of skillmake.ru — a Russian-language project about building real products with AI agents. The article you're reading is a translated case study from our production experience.