"Key Takeaways"
- Anthropic released Claude Fable 5.1 on September 1, 2026, as a generally available update to Fable 5.
- It is stronger on coding, knowledge work, and longer multi-step tasks, including research-style work, than Fable 5.
- Input and output prices stay $10 and $50 per million tokens. Cache reads drop 75% to $0.25 per million tokens.
- Claude Mythos 5.1 is the same model with looser safeguards, and it is not generally available.
Summarize this article with
Every few months a model ships with a new number on the end. Most of them are not worth rearranging your week.
Claude Fable 5 arrived in June. On September 1, 2026, Anthropic shipped Claude Fable 5.1. The name looks small. The useful part is not the decimal. It is that Anthropic is putting this model on the hard coding and knowledge work Fable 5 already did, and that long jobs got cheaper to run.
What is Claude Fable 5.1?
Think of it as the new Fable, not a side experiment. Anthropic describes Fable 5.1 and Claude Mythos 5.1 as the same underlying model with different safeguard levels. Fable 5.1 is the one most people can use. Mythos 5.1 stays inside trusted access programmes for vetted cybersecurity and life-sciences work.
If you call the API, the model ID is claude-fable-5-1. Anthropic also lists it on Amazon Bedrock, Google Cloud, Microsoft Foundry, and Claude Platform on AWS. Docs give it a 1 million token context window and up to 128,000 output tokens.
If you use Claude on the website, in Claude Code, or in Cowork, you can ignore the ID. What you need is simpler: Fable 5.1 is the Fable you get now. It defaults to High effort in Claude Code, and Medium effort in Cowork and on Claude.ai.
What changed from Fable 5?
- Stronger long-running coding and multi-step research, according to Anthropic's own evals.
- Better knowledge-work scores on documents, spreadsheets, and slides, again from Anthropic's published comparisons.
- More precise safeguards, so fewer everyday questions get blocked, especially some basic biology and defensive security queries.
- The same input and output prices as Fable 5, with much cheaper cache reads.
- Fable 5.1 can be used to identify software vulnerabilities. Exploit development and several other dual-use cyber tasks still get redirected to Opus models.
The numbers Anthropic published
| What they measured | Fable 5.1 | Fable 5 |
|---|---|---|
| Agentic scientific research (Terminal-Bench-Science 0.1) | 52.6% | 24.7% |
| Agentic coding (Terminal-Bench 4.0) | 55.8% | 42.0% |
| Cache reads (API token billing) | $0.25 per million tokens | $1.00 per million tokens |
| Typical workload cost vs Fable 5 | About 25% less | Baseline |
| Complex agentic workload cost vs Fable 5 | Up to about 45% less | Baseline |
Treat lab scores as a signal, not a promise
Those figures come from Anthropic's September 1, 2026 announcement and model docs. They are useful for direction. They are not a guarantee that your own coding session will feel twice as good.
The cost cut is more concrete for API and Claude Code users who already burn a lot of cached context. If you only send short chats on Claude.ai, you may notice the quality more than the price.
What can it actually do for normal users?
If you write, plan, or edit, the useful change is stamina. Anthropic's early-access partners said Fable 5.1 stayed easier to follow on long, multi-step work than Fable 5. That matters more than another headline benchmark if you have ever watched a model drift after twenty turns.
If you code, look at the longer sessions. Anthropic says Fable 5.1 is better at fixing root causes instead of papering over bugs, and that at Low or Medium effort it can match or beat Fable 5 while costing less.
If you research, Fable 5.1 is the public model. The more specialised science demos in the launch post, including some biology work, used Mythos 5.1 or extra tools. Do not assume the Fable you can open today will design proteins. Do assume it is stronger at reading a pile of documents and keeping a plan together.
Who should switch, and who can wait?
If Fable 5 already sits in your daily stack, and you pay for long coding or research runs, try 5.1. Cheaper cache reads plus a stronger default is a boring reason. It is also a good one.
If you are happy on Opus 5, ChatGPT, or another model, and you only need short answers, wait. Anthropic's own docs still tell many developers to start with Claude Opus 5, then move to Fable 5.1 when the harder reasoning jobs need it.
Enterprise buyers have a separate detail: Enterprise Frontier Safeguards, which store data in the customer's own cloud. Anthropic says that rolls out in phases later this fall. Until then, eligible customers can use Fable 5.1 with zero data retention. That is not a setting on a free personal chat.
Frequently Asked Questions
Is Claude Fable 5.1 the same as Claude Mythos 5.1?
Anthropic says they share the same model weights and differ in safeguards. Fable 5.1 is generally available. Mythos 5.1 is limited to trusted access programmes.
Did the price of Claude Fable go up?
No. Input and output prices stay $10 and $50 per million tokens, the same as Fable 5. Cache reads now cost 75% less, at $0.25 per million tokens.
Where can I use Claude Fable 5.1?
Anthropic says it is available on Claude's own products and on Amazon Web Services, Google Cloud, and Microsoft Azure. Developers call it with the ID claude-fable-5-1.
Should I leave Fable 5?
If you already use Fable 5 for coding or long documents, try 5.1. If you barely use Claude, leave your tools alone. A new decimal is not a reason to rearrange your week.
The Chai Takeaway
If Fable already does your long work, 5.1 is the update to try this week. If you only need short answers from another model, you can let this one pass. The best model is still the one that finishes your actual job, not the one with the newest number.
Explore More AI Updates