Plain-English Explainer
What is Claude Fable 5.1, and when should you use it?
A new name on the model list, explained without the launch-day noise.
You saw the name Claude Fable 5.1 and want to know what it is, who makes it, and whether it is the model you should be using. This page covers what it does, how it relates to Claude Mythos 5.1 and to Fable 5, where it sits next to Opus 5.5, Sonnet 5.5 and Haiku 5.5, and how to give it work that actually suits it.
Claude Fable 5.1 is a large language model made by Anthropic and released on September 1, 2026. It extends Claude Fable 5 and is aimed at demanding reasoning and long-horizon agentic work, including long-running coding, multistep research, and document, spreadsheet, and slide work. Anthropic's advice is to start with Claude Opus 5.5 for most workloads and move to Fable 5.1 when the task is harder or Opus 5.5 at higher effort still falls short.
Published
Free lesson · no account neededMeet ClaudeWhy Claude?6 minStart the free lessonKey Takeaways
- Claude Fable 5.1 is Anthropic's model for demanding reasoning and long-horizon agentic work. It was released on September 1, 2026 and is the successor to Claude Fable 5.
- Fable 5.1 and Mythos 5.1 are the same model with different safeguards. Fable 5.1 is generally available. Mythos 5.1 is limited to Anthropic's trusted access programs for cybersecurity and life sciences work.
- Anthropic's guidance is to start with Claude Opus 5.5 for most work. Reach for Fable 5.1 when the reasoning is demanding, the task runs long, or Opus 5.5 at higher effort still falls short on your own evals.
- Fable 5.1 has a 1M token context window, 128K max output tokens, text and image input, and always-on adaptive thinking that you steer with the effort setting.
- Anthropic says Fable 5.1 at Low or Medium effort matches or beats Fable 5's results, so you do not need maximum effort to see the improvement.
Claude Fable 5.1 at a glance
| Attribute | Claude Fable 5.1 |
|---|---|
| Made by | Anthropic |
| Released | September 1, 2026 |
| Relationship to earlier model | Extends Claude Fable 5 (released June 9, 2026), which is now a legacy model |
| Anthropic's positioning | Demanding reasoning and long-horizon agentic work |
| Strengths named by Anthropic | Long-running agentic coding, multistep research, and document, spreadsheet, and slide work |
| Context window | 1M tokens |
| Max output | 128K tokens |
| Input and output | Text and image input, text output |
| Thinking | Adaptive thinking, always on, steered with the effort setting |
| Default effort | High in the API and Claude Code, Medium in Claude Cowork and on Claude.ai |
| Comparative latency | Slower than Opus 5.5, Sonnet 5.5, and Haiku 5.5 |
| Reliable knowledge cutoff | June 2026 |
| API model ID | claude-fable-5-1 |
| Available on | Claude API, Amazon Bedrock, Google Cloud, Microsoft Foundry, and Claude Platform on AWS |
| Earliest retirement | Not sooner than September 1, 2027 |
Copy and Paste
Prompts You Can Use Today
Multistep research brief with a clear stopping rule
You are a research analyst preparing a briefing for a operations director who has 10 minutes to read. Topic: how mid-sized logistics companies are adopting route-optimization software. Work in stages. First list the five questions that would most change a decision. Then investigate each question using only the sources I have attached, quoting the passage you rely on. Finally write a one-page brief with a recommendation, the evidence for it, and a separate list of what the sources do not tell us. Mark any claim you could not support from the attachments as unsupported.
Swap the topic, audience, and reading time. Attach your own sources, because the 'only the attachments' rule is what keeps the brief honest.
Long document synthesis across many files
I am pasting twelve customer interview transcripts, each labeled with the customer's company size and role. Read all of them before answering. Identify the six themes that appear in at least three interviews, and for each theme give a one-sentence description, the interview labels that support it, and one verbatim quote. Then list any strong opinion held by only one customer that product leadership should still know about. Output as a table followed by a short paragraph on where the transcripts disagree with each other.
Change the number of files and the minimum count for a theme. The long context window is what makes reading everything at once practical.
Spreadsheet analysis with a built-in check
Here is a CSV export of our monthly support tickets with columns for date, category, priority, time to first response, and resolution status. Find the three most important trends over the period, explain each in plain language for a non-technical manager, and tell me which columns you used. Before you report any trend, verify it a second way, for example by recomputing it by category or by month, and say if the second check disagreed. End with five follow-up questions I should ask the support lead.
Replace the column list with your own. The 'verify a second way' instruction is the part worth keeping for any numeric work.
Slide deck from raw notes
Turn the meeting notes below into a 10-slide outline for a quarterly business review aimed at executives who have not seen the project. For each slide give a title that states the takeaway as a sentence, three bullets at most, and a speaker note of two or three sentences. Put the decision we need from the audience on slide 2, not at the end. Flag any slide where the notes are too thin to support the claim, and ask me for what is missing instead of filling the gap.
Adjust the slide count and audience. Keeping the 'ask me instead of filling the gap' line prevents confident slides built on thin notes.
Agentic coding task with scope and review gates
In this repository, migrate our date handling from the legacy helper module to the standard library across the whole codebase. Start by listing every file that imports the legacy module and propose a plan in the order you would change them. Do not begin editing until I approve the plan. After approval, work in small commits, run the existing test suite after each one, and stop and report if any test that passed before now fails. Summarize what changed, what you chose not to touch, and any risk I should review.
Replace the migration with your own refactor. The plan-then-approve gate matters most when the task will run for a long time.
Compare two options and argue against your own conclusion
I need to choose between building an internal reporting tool and buying a vendor product. I will paste our requirements, team size, and constraints. Give me a recommendation, then switch roles and write the strongest case against it, written by someone who wants the opposite outcome. Finish by telling me which single fact, if I could verify it this week, would most likely change your recommendation, and how I could check it.
Paste real requirements. The closing question turns the answer into a next step instead of a verdict.
What it does
What Claude Fable 5.1 is built for
Anthropic describes Fable 5.1 as one of its most advanced models for coding and knowledge work. The strengths it names are specific, and they share a theme: tasks that take many steps and long working sessions.
Long-running agentic coding
Work where Claude plans, edits many files, runs tests, and keeps going over a long session, such as a codebase-wide migration or a feature that touches several services.
Multistep research
Questions that need breaking apart, investigating in stages, and reassembling into a conclusion, rather than a single lookup and summary.
Document work
Reading long reports, contracts, transcripts, or policy documents and producing structured analysis, summaries, or rewrites.
Spreadsheet work
Analyzing tabular data, finding patterns, building the reasoning behind a number, and explaining it to someone who does not live in the spreadsheet.
Slide work
Turning notes and analysis into the structure of a presentation, with a point of view on each slide and speaker notes to match.
Demanding reasoning
Problems where the hard part is thinking carefully through constraints and trade-offs, not retrieving a fact.
Plain-English definition
What 'Claude Fable 5.1' actually means
Claude is the family of AI models made by Anthropic. Within that family, each model has a name and a version number. Fable is the model line Anthropic points to for its most demanding reasoning work, and 5.1 is the version. So Claude Fable 5.1 is the newer of two Fable releases, following Claude Fable 5, which Anthropic released on June 9, 2026. Fable 5.1 followed on September 1, 2026.
The model is a large language model, which means it was trained on a very large amount of text and learned to predict and generate language, code, and structured output. In practice that translates to a system you can talk to in plain English, hand documents and images to, and ask to reason, write, analyze, or act. Fable 5.1 takes text and images as input and produces text as output. It does not generate images or audio.
If you use Claude through the Claude app, through Claude Code, or through a tool built on the Claude API, the model you are talking to is one of a handful of options. Fable 5.1 is the one Anthropic points to for the hardest and longest tasks. It is not a separate product or a different kind of AI. It is a model you select, the same way you would pick a tool from a toolbox because of the job in front of you.
Fable, Mythos and safeguards
How Claude Fable 5.1 and Claude Mythos 5.1 relate
Anthropic introduced Fable 5.1 alongside a second name, Claude Mythos 5.1, and describes the two together as its most advanced models for coding and knowledge work. The relationship is simple once you hear it plainly: they are the same model with different levels of safeguards. There is no hidden second architecture and no capability gap you need to decode from the names.
Fable 5.1 is the generally available version. If you are signing up, building an application, or choosing a model in a product, Fable 5.1 is the one you can use. Mythos 5.1 is available only through Anthropic's trusted access programs. Its safeguards are designed for work in cybersecurity and the life sciences, fields where legitimate specialists sometimes need the model to engage with material that a general release is more cautious about.
The practical takeaway for most readers is that the name Mythos is not a better version you are missing out on. It is the same model configured for a narrower group of vetted users doing specialized work. If your job is writing, analysis, research, product, operations, or ordinary software engineering, Fable 5.1 is the model that applies to you.
Anthropic says the newer cybersecurity safeguards on Fable 5.1 flag fewer benign requests than before. That matters to anyone who has had a reasonable security question refused because it looked superficially like a risky one. A developer asking how a class of bug works, or a defender asking how to recognize a vulnerable pattern in their own code, should run into fewer false alarms.
Anthropic also says Fable 5.1 can be used to discover software vulnerabilities, though not to develop exploits for them. Finding a flaw in code so it can be fixed is within bounds. Writing a working exploit for that flaw is not something the generally available model will do. For security work, use it on systems you own or are authorized to test, and treat its findings as leads to verify, not as conclusions.
What changed
What is different from Claude Fable 5
Anthropic says Fable 5.1 is capable of much higher performance than Fable 5. The areas it highlights are stronger long-running agentic coding, multistep research, and document, spreadsheet, and slide work. Those are the same kinds of tasks Fable 5 was built for, so this is a step up in how well the model does the work, not a change in what the model is for.
One detail is easy to miss and useful to know. Anthropic says that at Low or Medium effort, Fable 5.1 achieves results similar to or better than Fable 5's. In other words, you do not need to run the new model at its maximum setting to see an improvement over the previous generation. If you were using Fable 5 and care about response time, you can often move to Fable 5.1 at a lower effort level and keep the quality you had.
Fable 5 is now a legacy model. Anthropic's documentation lists it with an Active (legacy) status, and it is no longer part of the current lineup, where Fable 5.1 is its successor. If you want the background on the earlier model, the Nightschool AI page on Fable 5 covers it, and this page focuses on what is current.
The lineup
Anthropic's current Claude models side by side
Anthropic's current lineup has four models. Each has a one-line job description, and the comparative latency runs from slower to fastest as you go down the list.
Claude Fable 5.1
For demanding reasoning and long-horizon agentic work. Latency is slower than the others, which is the trade for its depth.
Claude Opus 5.5
For long-running agentic coding and knowledge work. Latency is moderate. Anthropic says it performs at the level of Fable 5.1 on most work.
Claude Sonnet 5.5
The best combination of speed and intelligence in the lineup. Latency is fast, which suits everyday interactive work.
Claude Haiku 5.5
For high-volume, latency-sensitive tasks such as classification, extraction, and routing. It is the fastest model in the lineup.
Which model should I use
Which Claude model should you start with
Anthropic's own advice is direct: for most workloads, start with Claude Opus 5.5. Use Claude Fable 5.1 for demanding reasoning and long-horizon agentic work, or when evals on Opus 5.5 at higher effort still fall short. That is the whole decision rule, and it is more useful than it first looks, because it tells you to default to the model below the top one and move up only on evidence.
The reason this makes sense is that Anthropic says Opus 5.5, released on September 22, 2026, performs at the level of Fable 5.1 on most work. If the faster model is as good as the top model for your task, the slower model buys you nothing. Fable 5.1 earns its place on the tasks where that is not true: a hard multistep investigation, a very long agentic coding run, or a reasoning problem that Opus 5.5 keeps getting subtly wrong.
The phrase 'evals on Opus 5.5 at higher effort' is the practical part. An eval here just means a set of your own representative tasks with a way to judge the answers. Run them on Opus 5.5 first, raise the effort setting if the results are close but not good enough, and only if they still fall short move the same tasks to Fable 5.1. That sequence gives you a reason for the switch instead of a hunch.
For lighter jobs, go the other direction. Sonnet 5.5 is the best combination of speed and intelligence and handles a great deal of daily writing, summarizing, and question answering without a wait. Haiku 5.5 is the right pick when you are processing a large volume of items and each one needs a quick, simple decision, such as sorting support emails into categories or pulling a field out of a form. If you want a deeper look at the middle of the lineup, the Nightschool AI comparison of Sonnet and Opus walks through it.
Effort and thinking
How adaptive thinking and the effort setting work
Fable 5.1 uses adaptive thinking, and it is always on. That means the model decides how much reasoning a given request deserves before it answers, rather than you switching a thinking mode on or off. A simple question gets a short think. A hard one gets a long one.
You steer that behavior with the effort setting, not with an on/off toggle. Higher effort pushes the model toward deeper thinking, and lower effort pushes it toward faster, lighter answers. Anthropic documents High as the default effort for the model, and says it defaults to High in Claude Code and to Medium in Claude Cowork and on Claude.ai.
This gives you a practical dial. For a quick rewrite or a simple extraction, a lower effort level is faster and usually enough. For a research brief or a hard debugging session, High is the right place to be. And because Anthropic says Fable 5.1 at Low or Medium effort delivers results similar to or better than Fable 5's, the lower settings are not a downgrade to be embarrassed about.
Where to get it
Where you can use Claude Fable 5.1
For developers, the API model ID is claude-fable-5-1. The model is available on the Claude API, Amazon Bedrock, Google Cloud, Microsoft Foundry, and Claude Platform on AWS. That spread matters if your organization already standardizes on one cloud: you can usually reach the model from the platform your security and procurement teams already approve.
Anthropic has also committed that it will not retire Fable 5.1 sooner than September 1, 2027. If you are building something that will live for a while, that gives you a minimum window to plan around, so a model change does not arrive without notice.
For people who are not writing code, Fable 5.1 shows up through Anthropic's own products. The documentation mentions Claude Code, Claude Cowork, and Claude.ai as places where the model has default effort settings. Which models appear in the picker can depend on your plan and your product, so check the model selector in the app you use instead of assuming.
How to prompt it
How to write prompts for long, multistep work
A model built for long-horizon work rewards a different prompt style than a quick question does. Start by stating the end product: who will read it, how long it should be, and what format it should take. Then give the model the material it should rely on, and say plainly whether it may use anything beyond that. Most disappointing long outputs trace back to a missing audience or an unstated source rule.
Next, break the work into stages and ask for them in order. A prompt like 'first list the questions, then investigate each, then write the brief' produces a more reliable result than 'write me a brief,' because the model has a structure to follow and you have places to check. For anything that will run a long time, add a gate: ask for the plan first and tell it not to start until you approve.
Finally, ask the model to separate what it knows from what it is guessing. Instructions such as 'mark any claim you could not support from the attachments' or 'verify each trend a second way' cost you a sentence and save you from confident mistakes. The prompt examples on this page use all three habits. The Nightschool AI prompt engineering lesson goes further on each of them if you want to practice.
Honest limits
What Claude Fable 5.1 does not do for you
It is slower. Anthropic lists Fable 5.1's comparative latency as slower than Opus 5.5, Sonnet 5.5, and Haiku 5.5. The depth that makes it suitable for hard problems is the same thing that makes it a poor fit for quick back-and-forth, and for high-volume tasks where each item needs an instant answer.
It has a knowledge boundary. The reliable knowledge cutoff is June 2026, so events after that are outside what it reliably knows unless you give it the information or a tool to look it up. If your question depends on something recent, paste in the source.
It reads text and images and writes text, and nothing more. Input is text and images and output is text, so it will not produce images or audio. Its output can also be wrong, including on hard problems where it sounds sure. Check numbers, citations, and code before you rely on them, and keep a person responsible for any decision with real consequences in areas like legal, medical, financial, or safety-critical work.
And it is not automatically the right choice just because it is at the top of the list. Anthropic itself says Opus 5.5 performs at the level of Fable 5.1 on most work. If you cannot show that your task does better on Fable 5.1, the faster model is probably serving you well.
Learn to get real work out of Claude
Reading about a model and being good at using it are different things. The free Nightschool AI curriculum is hands-on from the first lesson, so you practice the prompts and habits on this page with real tasks instead of just reading about them.
Frequently Asked Questions
What is Claude Fable 5.1?
Claude Fable 5.1 is a large language model made by Anthropic and released on September 1, 2026. It extends Claude Fable 5 and is built for demanding reasoning and long-horizon agentic work, including long-running coding, multistep research, and document, spreadsheet, and slide work. It takes text and images as input, produces text as output, and has a 1M token context window.
When was Claude Fable 5.1 released?
Anthropic released Claude Fable 5.1 on September 1, 2026. It succeeded Claude Fable 5, which was released on June 9, 2026. Anthropic later released Claude Opus 5.5 on September 22, 2026, and says that model performs at the level of Fable 5.1 on most work.
What is the difference between Claude Fable 5.1 and Claude Mythos 5.1?
They are the same model with different levels of safeguards. Fable 5.1 is generally available to the public. Mythos 5.1 is available only through Anthropic's trusted access programs, with safeguards designed for work in cybersecurity and the life sciences. Mythos is not a more capable model, it is the same model configured for a narrower group of vetted users.
Should I use Claude Fable 5.1 or Claude Opus 5.5?
Anthropic advises starting with Claude Opus 5.5 for most workloads. Use Fable 5.1 for demanding reasoning and long-horizon agentic work, or when your own evals on Opus 5.5 at higher effort still fall short. Since Anthropic says Opus 5.5 performs at the level of Fable 5.1 on most work, moving up makes sense only when you can show your task needs it.
What is new in Claude Fable 5.1 compared with Fable 5?
Anthropic says Fable 5.1 is capable of much higher performance than Fable 5, with stronger long-running agentic coding, multistep research, and document, spreadsheet, and slide work. It also says that at Low or Medium effort, Fable 5.1 achieves results similar to or better than Fable 5's. Fable 5 is now a legacy model and no longer in the current lineup.
How big is the Claude Fable 5.1 context window?
Claude Fable 5.1 has a 1M token context window and a maximum output of 128K tokens. It accepts text and image input and returns text output. A large context window means you can give it many long documents at once, such as a set of interview transcripts or a large body of reports, and ask questions across all of them.
What is the API model ID for Claude Fable 5.1?
The API model ID is claude-fable-5-1. The model is available on the Claude API, Amazon Bedrock, Google Cloud, Microsoft Foundry, and Claude Platform on AWS. Anthropic has said it will not retire the model sooner than September 1, 2027, which gives developers a minimum window to plan around.
What does the effort setting do on Claude Fable 5.1?
Fable 5.1 uses adaptive thinking that is always on, so the model decides how much to reason about each request. The effort setting steers that depth. Higher effort leads to deeper thinking, lower effort to faster answers. The default is High, including in Claude Code, while Claude Cowork and Claude.ai default to Medium.
Can Claude Fable 5.1 find security vulnerabilities?
Anthropic says Fable 5.1 can be used to discover software vulnerabilities, though not to develop exploits for them. It also says the newer cybersecurity safeguards flag fewer benign requests than before. Use it on systems you own or are authorized to test, and verify any finding yourself before acting on it.
Is Claude Fable 5.1 faster or slower than other Claude models?
It is slower. Anthropic lists Fable 5.1's comparative latency as slower, Opus 5.5 as moderate, Sonnet 5.5 as fast, and Haiku 5.5 as the fastest. That trade buys depth on hard problems. For quick interactive work or high-volume tasks like classification, extraction, and routing, a faster model is usually the better choice.
Pick the right model, then get good at using it
The name on the list matters less than the habits you bring to it.