Claude Fable 5 now costs 5x the everyday AI models. After a month of real use, here are the only five marketing jobs where the premium pays — and the routing rule that protects your budget.
On July 7, the free ride ended. Claude Fable 5 — the most capable AI model anyone can currently buy — stopped being included in paid Claude plans and moved to metered billing at $10 per million input tokens and $50 per million output tokens. That makes it the single most expensive model on Anthropic’s price list: double the cost of their next model down, and roughly five times the cost of the workhorse model most of us use every day. Every launch post will tell you what Fable 5 can do. Almost nobody will tell you when you, specifically, should pay for it — and when reaching for it is just an expensive habit. I’ve been using Fable 5 in my own business since launch. Here’s the honest verdict, with real numbers.The one-sentence verdict
Fable 5 is a specialist you hire for your hardest jobs, not a daily driver. Route accordingly and it’s the best money you’ll spend on AI; use it by default and it’s the worst.Here’s why. For most of what online marketers do all day — emails, social posts, sales copy, rewrites — Fable 5 produces work you literally cannot tell apart from models costing a fifth as much. Where it separates, dramatically, is on a small number of big, tangled, multi-step jobs. Hand it those and it does things no cheaper model can. Hand it your daily emails and you’re lighting money on fire in $4 increments.
The math nobody puts on the sales page
A serious working session — feeding the model roughly 150,000 words of material and getting a 30,000-word analysis back — costs about $4.00 on Fable 5. The identical job on Sonnet 5, Anthropic’s everyday model, costs about $0.80. Read that both ways. If the job is “summarize some blog posts,” you just overpaid five-fold for the same summary. But if the job is “find the leak in my funnel that’s costing me $500 a month,” then $4 for a materially better answer is the cheapest consultant you will ever hire. The price is never the question. The question is whether the job is one where Fable’s answer is actually different.The pattern that decides everything
Buried in Anthropic’s own launch post is the line that matters more than every benchmark chart combined: “the longer and more complex the task, the larger Fable 5’s lead over our other models.” On short tasks, the gap between Fable and cheaper models is small — often invisible. On long, multi-source, multi-step tasks, the gap becomes enormous. That single pattern sorts every job in your business into “worth it” or “waste.”The five jobs where the premium pays
After a month of real use, these are the only five categories where I reach for Fable 5 without hesitating:1. Deep audience research in one pass
Cheaper models summarize audience material. Fable synthesizes it. Feed it hundreds of reviews, comments, and survey responses in one session and it surfaces the buyer segments, the objections in the market’s own words, and — most valuably — the places where the data contradicts what your copy assumes. Research that used to take a week of manual reading, done in one sitting.2. The full-funnel audit
This is the clearest case I’ve found where the premium pays. Funnel leaks live in the seams between assets — the ad promises one thing, the emails quietly drift to another. Finding those seams requires holding your entire funnel in mind at once. Cheaper models audit page by page and miss them. Fable reads the whole thing as one system. One found leak pays for a year of this.3. Competitor gap mapping
Feed it every competitor’s sales page, email sequence, pricing, and — crucially — their customers’ complaints, then ask for the gap nobody is filling. The output is positioning you can defend, with evidence attached.4. Building tools you can sell or give away
Fable one-shots working software — calculators, quiz funnels, interactive planners — that took earlier models a hundred prompts of correction. A working branded tool is a lead magnet or bonus that no PDF competes with.5. Turning your archive into a product
If you’ve been publishing for years, you’re sitting on a product you haven’t compiled. Give Fable years of newsletters or transcripts and it finds the teachable framework that already exists in them and builds the course outline around your own material. Cheaper models lose the thread at this volume and start padding. Fable doesn’t.The routing rule that protects your budget
Here’s the rule of thumb I now run everything through:If the whole job fits in one screen of prompt and one screen of answer, use the cheap model. If you’re feeding the model more material than you could read in an hour and asking for judgment across all of it, that’s a Fable job.The trap is psychological, not technical. “Most powerful” quietly becomes “default,” and the default runs all day on tasks where it adds nothing. That’s how a premium tool becomes a premium waste — a few dollars at a time, invisibly, all month.
What I haven’t covered here
This post gives you the framework. The full report goes considerably further, including:- The ten complete deep-work prompts I use for the five jobs above — avatar excavation, funnel teardown, bonus-stack architecture, list reactivation, launch debriefs, pricing analysis and more, ready to fill in and run
- The fine print the hype skips — including a safety-filter quirk that means you’re sometimes paying Fable prices for a different model’s answers, and exactly when that happens
- The everyday task list where Fable is a proven waste (this section saves most readers more money than anything else)
- Who should reach for it and who should skip it entirely for now
- The One-Job Test — a simple experiment that settles the whole question on your own business, this week, for a few dollars, so you’re deciding from evidence instead of launch hype
Pricing and product facts are from Anthropic’s published announcements as of July 7, 2026. AI pricing changes fast — confirm current rates before making budget decisions. No hype, no tech waffle: that’s the whole point of Now I Get AI.
