Claude Fable 5 is the most capable model Anthropic has released. Buying it is still a procurement decision.

The sequence matters, because the entire risk lesson is compressed into a single month.

Anthropic released Claude Fable 5 on 9 June 2026, alongside a companion model, Mythos 5. They share the same underlying model, though Fable 5 carries additional safety classifiers that make it suitable for general release.

Three days later, on 12 June, the US government issued an export control directive on national security grounds. It prohibited access by any foreign national, whether inside or outside the United States. Anthropic had no reliable way to verify nationality in real time, so it deactivated Fable 5 and Mythos 5 for every customer at once.

The trigger was a reported jailbreak. Amazon researchers discovered a way to prompt Fable 5 so that it identified software vulnerabilities, and in one instance produced code demonstrating how a vulnerability could be exploited. Anthropic examined the claim and confirmed that the same vulnerabilities could be identified by numerous less capable models, its own Claude Opus 4.8 among them. Every model it tested reproduced the identical demonstration.

The export controls were lifted on 30 June, and Fable 5 returned worldwide on 1 July with an improved classifier that blocks the reported technique in over 99% of cases.

Total the damage. A commercial tool relied upon by hundreds of millions of people went dark for nineteen days, without notice, over a directive the manufacturer itself openly disputed. Keep that number in mind, because it is the entire point of this page.

KEY POINT
The outage was not a configuration problem you could have onboarded your way around. It was a government order the supplier had to obey overnight. Nothing on your side would have kept the tool available.

Now the economics. Fable 5 costs 10 US dollars per million input tokens and 50 US dollars per million output tokens, which is twice the price of Opus 4.8, the model immediately below it. Until 11:59 pm PT on 12 July, it remains included on Pro, Max, Team and selected Enterprise plans, for up to half of your weekly usage allowance. Beyond that deadline, you pay for every use.

Set the enthusiasm aside and three procurement tasks align with what this model is actually good at.

Long documentation

Fable 5 holds up to a million tokens of context simultaneously. Translated into practical terms, that is an entire stack of contracts, a complete policy set, or a data room absorbed in a single pass without losing the thread. For contract and document review, that reach is the capability that genuinely matters.

Prolonged, unattended work

Anthropic's own description is that the model sustains its concentration across very long tasks and refines its output using its own notes. A continuous audit that would ordinarily proceed in fits and starts is precisely the sort of assignment it suits.

Reading scanned paperwork

Fable 5 is state of the art on vision, including extracting precise figures from detailed documents. Most supplier paperwork in a smaller business is not tidy digital text. It is scanned PDFs, photographed invoices, and handwritten forms. A model that reads those accurately recovers genuine hours.

Notice the pattern: these are substantial, document-heavy undertakings, and that observation determines the entire question.

Two limitations matter considerably more than any benchmark.

The first is a condition you cannot disable. Every prompt and every response on Fable 5 is retained by Anthropic for 30 days, across every platform where the model operates. It exists to support their safety work, and for most consumer subscriptions that arrangement was already in place. The complication for businesses arrives next. If your organisation holds a zero data retention agreement, that agreement does not extend to Fable 5. To use the model at all, you must switch 30-day retention on for those workspaces.

Reread that if you handle confidential supplier or client information. A non-disclosure agreement you signed with a client does not dissolve simply because a model is capable. Paste a confidential tender, a priced submission, or personal data into Fable 5, and it resides with a third party for 30 days. That may be entirely acceptable. It may equally breach an obligation you have already made. The discipline is to decide deliberately, not accidentally.

KEY POINT
If you are bound by a zero-retention or confidentiality obligation, treat Fable 5 as a third party who keeps a copy for a month. Clear the data before you clear the tool.

The second limitation is quieter. Fable 5's safety classifiers are deliberately conservative. When a request resembles cybersecurity, biology, chemistry, or a handful of other domains, the model redirects the answer to Opus 4.8 instead, and Anthropic has tuned this to trigger on some harmless requests as well. For routine procurement work you will encounter it rarely. It is nonetheless worth knowing the tool occasionally declines for reasons entirely unrelated to you.

A third limitation is more obvious. A capable model still does not understand your business, your market, or what you genuinely require. It drafts. It does not decide. That distinction is the whole of good procurement.

The sticker price is the straightforward part. Fable 5 costs 10 US dollars per million input tokens and 50 per million output, which is double Opus 4.8 and more than three times the price of Sonnet 5, the mid-tier model. For intensive use, that accumulates rapidly.

Claude Fable 5
$10 / $50
input / output per million tokens

Claude Opus 4.8
$5 / $25
input / output per million tokens

Claude Sonnet 5
$3 / $15
input / output per million tokens

The larger cost never appears on the invoice. It is the consequence of constructing a way of working around a single supplier, and then watching that supplier disappear for nineteen days without warning.

We say this to clients about every supplier, and it applies precisely here. The moment a single provider becomes load-bearing within your business, their problem becomes your problem. A directive issued on the other side of the world removed this tool from everyone simultaneously, overnight. Had your month-end audit or your tender evaluation been living inside it, you would have been stranded, with nothing to telephone and nobody to chase.

KEY POINT
Price is what you pay per token. Cost is what happens to you when the single tool you built around switches off. Budget for both.

Opinions are like bellybuttons. Everyone has one about which AI you ought to use, and most of those opinions are selling something. So set them aside, and apply the same examination we would apply to any purchase.

Buy what you need, not what you're sold. Work through five questions.

1. What's the job? Name the actual task, not the tool. "Read forty contracts and list every automatic renewal" is a job. "Get Fable 5" is not.
2. Does a cheaper option already do it? For most drafting, summarising, and first-pass review, a mid-tier model you already pay for is sufficient. Fable 5's advantage emerges on long, complex work, not on a routine email.
3. How often will you do it? An annual audit does not justify a permanent bill. A daily task might.
4. What's the consequence of it being offline? If the work cannot tolerate a nineteen-day absence, do not make it dependent on one tool that can.
5. What are you feeding it? If the honest answer is confidential client or supplier data, the 30-day retention condition settles this before price does.

KEY TAKEAWAY
If you cannot state plainly what the job is, how often you will do it, and what happens if the tool vanishes, you are not buying a capability. You are buying a subscription and hoping. Hope is not a procurement strategy.

A short list of substantial jobs makes Fable 5 worth the expense, and all of them are document-heavy and genuinely difficult to accomplish any other way.

• A full contract register audit. Every active contract read in a single pass, with renewal dates, price mechanisms, and exit rights extracted and compared.
• Multi-bidder tender evaluation at scale. A dozen lengthy submissions assessed against identical criteria, without fatigue distorting your judgement by submission nine.
• An M&A or due-diligence data room. Hundreds of documents that must be read, cross-referenced, and summarised against a deadline.
• Building a governance framework from a blank page. Drafting the thresholds, roles, and review cycles you can then refine into your own.

We have not yet run a published D1 Advisory engagement on Fable 5, so we are not going to offer you a savings figure we cannot substantiate. What we can describe is the shape of the work. If your task resembles the four above, the model's reach and stamina accomplish something a cheaper tool cannot. If it does not, you are paying frontier prices for work a mid-tier model completes just as capably.

KEY POINT
Fable 5 repays the investment on the major undertakings, not the daily grind. Match the tool to the scale of the job.

At present, Fable 5 is included in your paid subscription, up to half of your weekly usage, until 11:59 pm PT on 12 July. That is effectively a free trial you never had to request. Treat it as one.

Do not commit to paying on 13 July because a headline instructed you to. Instead, run your genuine work through it this week. Hand it the substantial, awkward job you have been postponing. Establish whether the output justifies the price. Then decide from evidence, rather than from a fear of missing out.

A trial delivers the most value when you test it on something that matters. Choose one real task. The seven prompts below give you somewhere to start.

These seven put Fable 5 to work on assignments a business owner genuinely has. Copy, paste, and substitute your own detail. One caution first, and it is the important one.

BEFORE YOU PASTE
Do not feed it anything you are not permitted to share. Remember the 30-day retention condition from earlier. Remove names, prices, and personal data, or substitute rounded figures and initials. Test the tool on a redacted version first.

If all of this has left you wishing someone would simply tell you whether to pay for the thing, trust that instinct. The answer depends on your contracts, your data, and how frequently you would genuinely use it. That is a conversation, not a checkout.

Book a discovery call with D1 Advisory. We will examine the actual jobs you would point it at, what you are permitted to feed it, and whether a cheaper tool performs the same work for less. We will not tell you AI will rescue your business, and we will not say “disruptive” once.

Buy what you need. Not what you're sold. Even when what you're being sold is remarkably, genuinely clever.