Published September 2, 2026
On September 2 we moved the Claude Fable mode on Secret Chat AI from Claude Fable 5 to Claude Fable 5.1. If you already use that mode, the menu row got a new name and nothing else you can see moved. The answers now come from Anthropic's newest top model, and a long conversation on it costs you less on the follow-ups.
What Claude Fable 5.1 Is
Fable 5.1 is Anthropic's most capable generally available model, released on September 1. It succeeds Claude Fable 5 in the same tier, which still sits above Claude Opus, and it's the same model as Claude Mythos 5.1, which Anthropic hands only to vetted organizations through its trusted access programs. The difference is the safeguards: Fable is the public version, with the stricter ones around cybersecurity and biology. Anthropic's own page is at anthropic.com/claude/fable.
The shape of the model hasn't changed since Fable 5. It reasons before every answer, and you can't turn that off. It reads up to a million tokens of context. It's built for the work where the other tiers start to fray: long chains of reasoning, dense documents, code with a bug three layers down, decisions with a dozen constraints pulling in different directions. What Anthropic says got better in 5.1 is long-running coding, multi-step research, and work on documents, spreadsheets and slides. It also says the safeguards get in the way less: by its own figures, 85% fewer interventions on ordinary biology and medical questions, and around 60% fewer per session on cybersecurity topics. We haven't measured either claim ourselves.
What Actually Changed
Here is the honest version. On the wire, almost nothing.
The request our gateway sends to Anthropic for Fable 5.1 is the request it sent for Fable 5: same endpoint, same context window, same output ceiling, same always-on reasoning. That's convenient and slightly dangerous at once. A half-finished upgrade would keep answering, on the old model, and nothing would fail. So the swap is pinned by tests that fail the moment the model behind that menu row, the price we charge for it and the name on the reply line disagree with each other. Boring, and the kind of boring that matters. Anthropic does list three breaking changes for developers moving from Fable 5, all about forced tool calls and reusing the model's thinking across turns. Our gateway does none of those things, so none of them reached us.
One thing did move, and it's good news for long conversations. Every time you send a message, the model re-reads the whole chat so far. Anthropic lets a gateway cache that earlier part rather than pay full price to re-read it, and ours has always done that on Claude. With 5.1, Anthropic cut the price of a cached re-read to a quarter of what it was on Fable 5. In credits: when your next message follows within a few minutes, which is how most chats go, a reply deep into a long Fable conversation now costs less than it did last week, and the further in you are, the more it shows. The first message costs what it did. A new question after a long pause costs what it did.
The other thing to know isn't new, but it's worth saying. Fable thinks before it answers, and on a hard question it thinks for a while. Anthropic itself lists it as the slowest model in the lineup. A spinner that runs longer than you're used to on Opus is usually the model thinking, not the app hanging.
Smart Agentic Claude Still Runs on Opus
Nothing changed here either, and we'd do it again. Pick Claude and leave it on Smart Agentic, and a router reads each message and decides how much work it deserves: a fast model for chat, web search for fresh facts, and Claude Opus when the question is hard. Fable isn't in that rotation.
The reason is cost. A Fable answer costs about twice the credits of the same answer from Opus, and the router fires on every message you send, including the ones that only need a quick reply. Put Fable in the rotation and everyday chats get twice as expensive without getting noticeably better. Opus is very good. For most questions the gap between it and Fable is small, and the gap in credits isn't. So Fable stays a mode you choose yourself.
How to Choose Claude Fable 5.1
- Open the app and pick Claude in the model picker.
- Open the mode menu next to the message box. Below Smart Agentic, Fast Chat, Web Search and Research, the last row is Claude Fable 5.1, with a star icon. Pick it.
It sits last on purpose. Every other row on that menu is a step on one ladder, from a quick reply to a long researched answer. Fable isn't a step on that ladder. It's a different, dearer model you pick deliberately, so it sits apart from the rows you normally choose between.
From then on, every message in that chat goes straight to Fable 5.1. It's web-grounded on Secret Chat: it can search the web and read pages while it works, so its reasoning runs on current information rather than on training data alone. Switch the mode back to Smart Agentic whenever you like and the router takes over again, in the same chat, with the history intact. Under each reply the app names the model that wrote it and the credits it spent, so you'll see "Claude Fable 5.1" and the price of that answer right there.
A prompt that gets the most out of it:
I have a genuinely difficult problem and I want your deepest reasoning. Take your time, think it through step by step, consider the approaches you would reject and why, and only then give me your best answer. Here's the problem:
When It's Worth It
Our rule: if you never ask anything harder than an email draft or a quick fact, skip this mode entirely. Smart Agentic already escalates to Opus when a question needs it, and that's the right level for most work.
Reach for Fable when the question is the kind you'd pay a specialist for. A long contract or technical spec you need pulled apart. A multi-step problem where the earlier tiers lose the thread halfway. A piece of analysis you'll act on. One workflow our users already run: ask Smart Agentic first, and if the answer to a hard question falls short, switch the same chat to Fable 5.1 and ask again. Two answers side by side tell you fast whether the mode earns its credits for your kind of work.
That comparison has a limit. Two Claude tiers agreeing tells you less than it looks like, because they come from the same lab and mostly the same training. Comparing across companies is what the AI Council does: one question to several models from different companies at once, and a referee that maps where they disagree, claim by claim. Fable isn't on a council panel, and that's by design. A council seats each family's mid tier, because its value is independent companies disagreeing, not each answer being the strongest available. Depth from one model is Fable's job. Whether several companies agree is the council's. You can also paste a Fable answer into a council and let the panel check it.
One Privacy Fact We'd Rather Tell You Than Have You Find
Anthropic classes the Fable tier as what it calls a Covered Model. Requests to it are kept on Anthropic's side for 30 days for safety monitoring, and zero data retention is off the table for it unless Anthropic expressly authorizes a customer. That was true of Fable 5 and it stays true of 5.1. We don't get to change it, and we'd rather say it here than leave you to find it in the fine print. Our longer piece on why zero data retention is harder than it sounds covers the same rule across providers.
What Secret Chat changes is whose record that is. Your prompt reaches Anthropic through our gateway, under our credentials, from our servers. The record on Anthropic's side carries our gateway's address, not your name, your account or your IP. You use the model as a stranger. Secret Chat AI removes you from your queries; it does not remove the data from your messages. Write your own name or your client's case number into a prompt and it's still sitting there in the prompt, on Fable as on any model. Your chat history lives only in your browser, and a prompt exists on our side only for as long as it takes to fetch your answer.
Claude Fable 5.1 is on Secret Chat now. Pick Claude, pick the last row in the mode menu, and give it the problem you've been putting off. The first draft of this post came out of Fable 5.1 itself, which is either a good demo or a conflict of interest. We'll let you judge.
Frequently Asked Questions
- What is Claude Fable 5.1?
Claude Fable 5.1 is Anthropic's most capable generally available model, released on September 1, 2026, and the successor to Claude Fable 5 in the tier above Claude Opus. It is the same model as Claude Mythos 5.1, which Anthropic offers only to vetted organizations; Fable is the public version with stricter safeguards. It always reasons before answering and reads up to a million tokens of context.
- How do I switch to Claude Fable 5.1 on Secret Chat?
Open the Secret Chat app, select the Claude model, open the mode menu next to the message box and pick the last row, "Claude Fable 5.1", marked with a star icon. To go back to automatic routing, switch the mode back to Smart Agentic in the same chat.
- What changed from Claude Fable 5 to 5.1 on Secret Chat?
The mode now runs Fable 5.1 and is named after it. The request shape is unchanged, so nothing about how you use it moved. What did change is that re-reading a cached conversation is now much cheaper on Anthropic's side, so replies deep into a long Fable chat cost fewer credits when your messages follow each other within a few minutes.
- Why isn't Fable part of Smart Agentic Claude?
Because a Fable answer costs about twice the credits of the same answer from Claude Opus, and the Smart Agentic router runs on every message, including ones that only need a quick reply. Keeping Opus as the router's top model keeps everyday chats affordable. Fable stays available as a mode you choose on purpose.
- Does Anthropic keep prompts sent to Claude Fable 5.1?
Yes. Anthropic classes the Fable tier as a Covered Model, keeps requests to it for 30 days for safety monitoring, and offers zero data retention on it only to customers it expressly authorizes. Through Secret Chat those requests arrive under our gateway's credentials and address, not your name, account or IP. That protects who you are, not what you type: anything identifying you write into a prompt is still in the prompt.
- Can Claude Fable 5.1 search the web on Secret Chat?
Yes. On Secret Chat, Fable is web-grounded: it can search the web and read pages while working on your request, so its answers combine deep reasoning with current information.
- Can I put Fable on an AI Council panel?
No. The AI Council seats each family's mid tier rather than its flagship, because a council is evidence when models from different companies disagree, not because each answer is the strongest obtainable. Use Fable when you want one model's deepest work on a hard problem, and a council when you want to know whether several companies agree. You can paste a Fable answer into a council for the panel to check.