Fable 5 Is The Legacy Model. Fable 5.1 Costs Exactly The Same.

A card went up listing “15 insane things you can do with Claude Fable 5 (that 99% of people don’t know)” — a wheel of fifteen spokes radiating out from one model name in the middle. The fifteen are a reasonable list. Several of them are things the model genuinely does well.

The problem is the name in the middle of the wheel. Claude Fable 5 is a legacy model. Anthropic’s own documentation says so in a banner at the top of its page. The current Fable model is Fable 5.1, it costs exactly the same, holds exactly the same context, produces exactly the same maximum output — and knows five more months of the world.

So this is not a post about whether the fifteen things are possible. It is about the two numbers on the card that are quietly behind the documentation, and what you should do about each one.

The name in the middle of the wheel is the previous model

Here is the page for Claude Fable 5, quoted exactly as it opens:

“Legacy. Released June 9, 2026. Although Claude Fable 5 is still available, you should consider migrating to Claude Fable 5.1.”

“Still available” matters. Nothing on the card is broken and nothing has been switched off — its retirement date is no sooner than 9 June 2027. But put the two models side by side and the case for naming the older one disappears.

Read that table as a buyer rather than a reader. There is no trade-off here. Usually “newer model” means a decision — more capable but slower, or better but dearer. Between these two there is no decision to make. Same money, same window, same ceiling, five months more world. If any of the fifteen things on the card touch something that happened in the first half of 2026, the older model is the one that will confidently tell you the stale version.

And there is a practical reason this matters more than it sounds. Nine of the fifteen spokes are research-shaped: summarise long research, prepare strategy documents, analyse financial data, improve legal documents, build training materials. Those are exactly the jobs where a knowledge cutoff is the binding constraint, because the model’s answer is only as current as the last thing it read.

The headline number on the card is low by a factor of six

The caption’s most concrete brag is “analyze 100-page PDFs”. It is meant to sound like a ceiling you would be impressed by. It is a ceiling — just not this model’s.

Anthropic’s PDF support page gives the limit as a single line with a condition attached:

“Maximum pages per request: 600 (100 when the request’s context window is under 1M tokens).”

Fable 5 has a 1M-token context window. So does Fable 5.1, and so does every other current Claude model. On the model the card is advertising, the documented page limit is six hundred, and 100 is the figure that applies when you are not on a million-token window.

That said, pages are the wrong unit to plan with, and this is the genuinely useful part. The docs also state the cost per page: “Each page typically uses 1,500–3,000 tokens per page depending on content density.” Two ceilings therefore apply at once, and for a dense document the token one arrives first.

So the honest version of the brag is this: 100 pages is not the limit, it is the comfortable zone. You have roughly three times that before a dense document starts crowding out everything else in the conversation, and the docs tell you what to do when you get there — “Split large PDFs into chunks when needed.”

Diagram: the fifteen spokes, sorted by where the work actually happens

The caption says all fifteen happen “inside one chat window”. The card itself disagrees in four places, and it disagrees in writing — two tiles name Claude Code, and the tile text names three products made by other companies. Sorting the fifteen by what you need open in front of you is more useful than ranking them.

Credit where it is due on that bottom row. A lot of cards in this genre imply one subscription does everything. This one writes “Combine ChatGPT Image, Seedance, and Claude Code to build sites” directly on the tile, which is a rival’s product named on a card about Claude. That is the opposite of overselling. HyperFrames, named on tile 11, is a real open-source renderer that Claude Code drives — the tile is accurate.

It is also consistent with something we found when we went through ten logo prompts written for Claude: the model does not make pictures. If a card about Claude needs an image, it has to borrow one from somewhere, and this card says where.

The footer deserves a mention too

In small type at the bottom of the card:

“Ideas for what AI can help with. Results vary and AI output should be reviewed by a person, especially for legal, financial and code work.”

That sentence names the exact three categories on the card where an unreviewed answer costs real money — tiles 12, 10 and 04. A card that lists “Improve Legal Documents” among fifteen insane things and then tells you to have a person read the output is being more careful than its own headline.

What to actually do with this card

1. Use the current model, not the one in the headline. Fable 5.1 over Fable 5. Same bill, newer knowledge. If you are picking in an app rather than by model ID, pick whatever the app calls its current top model.
2. Stop worrying about 100 pages. Plan in tokens instead: roughly 1,500–3,000 per page. Split when the document would eat most of the window, not when it crosses a page count.
3. Read the tile before you believe the caption. Four of the fifteen need something that is not a chat window, and the tiles say so plainly.
4. Take the footer at its word on three of them. Legal, financial and code output gets read by a person. That is the card’s advice, not ours.

Four honest caveats

1. “Legacy” is not “broken”. Fable 5 works, is supported, and will not retire before 9 June 2027. Nothing on the card fails because of the version number. The point is that the newer model costs the same, so there is nothing to weigh up.
2. We did not run the fifteen. Everything above is read off the card and checked against published documentation. Whether tile 11 produces a video you would actually publish is a different question from whether the tool exists.
3. Specs change. Prices, context windows, page limits and model line-ups all moved more than once this year. Everything here reflects the documentation as of October 2026; check the model page before you commit a budget to it.
4. The page limit has a condition attached and we may be reading the common case. The documented 600 applies to requests on a 1M-token context window, which currently means every Claude model; a product or platform that caps the window lower would land back on 100. The honest statement is that 100 is not the model’s limit, not that 600 is guaranteed for every route into it.

The one-line version

Both of this card’s hard numbers are one step behind the documentation: the model in the middle of the wheel is the legacy version of a model whose current release costs identically, and the “100-page PDF” headline is the limit for a smaller context window than the one that model actually has.

The fifteen things are fine. Do them on 5.1, and stop counting pages.