Claude Fable 5, the best model you can't afford to keep
The most powerful (and expensive) model Anthropic has ever released is back online, so what does it offer to marketers?
Fable 5 has had a strange few weeks. It launched June 9 as the most capable model Anthropic has ever released to the public. On June 12, the US government hit it with an export-control directive over a reported jailbreak technique, and Anthropic pulled access for everyone worldwide.
Fast forward to two days back, when Fable 5 came back online with a new deadline: included for up to 50% of your weekly usage limits on Pro, Max, Team, and select Enterprise plans through July 7. After that, it's usage credits: $10 per million tokens in, $50 per million out. In a sense, it bills you by the word, and it's not cheap.
Is it even worth paying for, or should you stick to Opus for now?
First, what kind of tool is it?
What is a model this expensive for, if you’re a marketer and not a venture-funded code factory?
It’d be a mistake to treat Fable like “Opus, but better.” It’s not built to be a smarter version of the same job, but built for a different kind of job entirely. Give it a huge pile of source material, let it work through the whole thing on its own without you checking in, and it hands back something finished.
It can actually work with a massive amount of information at once — not just skim it, but reason across all of it. And it plans out multi-step work itself, catching its own mistakes along the way, so you’re not babysitting it through each step.
You describe what you want, then leave it alone. The catch: it costs about twice as much as Opus and burns through tokens twice as fast, so it only pays for itself when the job is big enough to need all that.
Three marketing jobs I’d point it at
1. Positioning from raw evidence. Let’s take B2B SaaS companies as an example. Feed it competitor pages, G2 reviews, your own site, and real sales-call transcripts, and let it work in phases. First a teardown of every competitor’s homepage, pricing, and “vs you” attack page: verbatim quotes, links required, no hand-waving. Then an audit of your own messaging the same way. Then a self-critique pass, where it attacks its own draft positioning and kills anything that contradicts your actual reviews.
What lands at the end is a full framework: a positioning statement, message pillars with proof, rewritten homepage copy, objection scripts for sales, and a flagged list of problems that need a business decision, not better copy.
The APAC bonus: point it at your customers’ transcripts and it works from the vocabulary your buyers in specific markets use. It’s the kind of job that eats a strategist’s afternoon across a dozen tabs, done in one pass.
2. A full-site messaging audit from one prompt. Point it at your entire website and ask where the value prop drifts page to page, where the audience silently shifts from founders to enterprise and back, where a case-study page is still wearing a “start free trial” button, where one page’s claim quietly contradicts another’s.
It crawls the lot and returns a single prioritised report: biggest gaps up top, a page-by-page breakdown with the offending copy quoted, and a fix list ordered by likely traffic and conversion impact.
Opus can do a slice; Fable holds the whole site at once, which is the entire point. If your messaging has quietly fractured across landing pages over two years of “quick edits” nobody documented, this is what finds it.
3. Building a real, working thing. Not just a mockup, but a fully functioning, interactive tool from a prompt or two, with the logic, the UI, and deploy-ready code handled in one autonomous run.
The example worth stealing is a “Spotify Wrapped” for your customers: a personalised year-in-review showing how each one used your product, with nudges to get more out of it. But the shape generalises: an interactive ROI calculator for your pricing page, a self-scoring assessment or quiz as a lead magnet, a report generator that turns someone’s inputs into a tailored PDF.
It’s the sort of asset that used to mean a developer, a designer, and a sprint, now sketched and built in an afternoon. You’ll still definitely want a human QA pass before it goes live, but the distance from “idea” to “working” all but collapses.
The tell across all three: if a job would take a sharp person half a day across multiple sources, and the output is something more complex than a doc, that’s a Fable job.
What it’s not for
The drafting, the campaign angles, the ad variants, the email sequences, the rewrite-this-paragraph — the work that actually fills a marketing week — runs fine on the models you already have, and probably runs better. Newer and pricier hasn’t moved the needle on everyday marketing work for a couple of cycles now. Save Fable for the half-day heavy lift, not the Monday-morning email.
The cost underneath the price
Here’s the part underneath all of this. Every Claude model before Fable was just included in your subscription — you paid your monthly fee and used it. Fable breaks that pattern. Once your included usage runs out, you start paying per word, through the API.
And that bill is bigger than it looks. SemiAnalysis — a firm that specializes in figuring out what AI companies actually spend to run these models — has spent years digging into the real cost of serving something like Fable: the chips, the electricity, the data centers, all of it. Their finding is blunt: the price you’re quoted is a business decision layered on top of a much bigger real cost, and running the most powerful models at scale is expensive in a way that doesn’t show up on your bill until the free ride ends.
To be fair, it’s not all bad news — AI in general keeps getting cheaper. Older, less powerful models drop in price by roughly 10x a year as the tech improves. But that’s the cheaper, “good enough” tier. The frontier — Fable-level stuff — sits at the very top, priced close to what it actually costs, because nobody’s covering the difference for you there yet.
Which is why cost matters more for this tool than anything else we’ve covered. That’s fine if you’re a well-funded US startup that’ll never really feel the bill. It’s riskier if you’re a marketer in Jakarta or Manila or Ho Chi Minh City, working in a currency that isn’t the dollar, where $50 per million tokens turns into real money fast.
The rule that’s held all year still holds: don’t build your workflow around something you can’t afford once the discount ends. The win isn’t having access to the best model. It’s knowing exactly which slice of your work is actually worth paying for it.
The AI;DR
Elsewhere in the AIverse
AI hardware gets a serious new player. Etched came out of stealth with $5B in valuation and $1B in signed contracts for chips built to run AI cheaper than general GPUs, shipping this summer. If this delivers, it’s the biggest lever for bringing prices like Fable’s back down.
A new AI browser wants to just handle things. Aside, a YC-backed browser, claims #1 on three browser-agent benchmarks and shows itself cancelling unused subscriptions in its launch video. The benchmarks are self-reported, so treat with caution — but browsers are clearly the next agent battleground.
Gemini now touches your Mac’s actual files. Spark can sort a Downloads folder or turn local invoices into a spreadsheet, with phone-to-Mac task handoff coming soon. It’s beta, US-only, and gated behind $99/month Google AI Ultra — but every major lab is clearly racing toward agents that act on your desktop, not just chat.