Claude Fable is Anthropic’s public version of its Mythos-class models: the same very large model as Claude Mythos, released with safety classifiers so anyone can use it. The current version, Claude Fable 5.1, launched on 1 September 2026 and costs $10 per million input tokens and $50 per million output tokens.
Mythos, the same model with some safeguards lifted, is reserved for vetted US organisations. Fable is included on Max plans and Premium seats; Pro users pay with usage credits. Since 22 September, the cheaper Claude Opus 5.5 has overtaken it on most published tests. All details are as of 24 September 2026.
What is Claude Fable?
Claude Fable is Anthropic’s premium general-access model family and the public face of its Mythos-class models. Anthropic describes Fable 5 as a Mythos-class model “made safe for general use”: it runs classifiers that hand flagged cybersecurity and biology requests to older Claude models. Fable 5.1, released on 1 September 2026, is the current version.
Fable is available on claude.ai, in Claude Code, and through the Claude API, Amazon Bedrock, Vertex AI and Microsoft Foundry.
In the Fable 5 and Mythos 5 launch announcement, Anthropic quotes Stripe saying it “compressed months of engineering into days” (Anthropic-reported). Among other generative AI tools, treat Fable as the premium tier: capable, slow and priced to match.
When was Claude Fable released?
Claude Fable 5 launched on 9 June 2026 alongside Mythos 5, and Fable 5.1 followed on 1 September 2026. In between, US export controls forced Anthropic to suspend all access from 12 June until 1 July. Both models grew out of Claude Mythos Preview, which Anthropic announced in April 2026 for a closed group of partners.
Read more: What Is Claude? What It Does, What It Costs, What It Cannot Do
| Date (2026) | What happened |
|---|---|
| 7 April | Mythos Preview announced for Project Glasswing partners only |
| 9 June | Fable 5 and Mythos 5 launch |
| 12 June | US export controls; all access suspended |
| 30 June | Export controls lifted |
| 1 July | Access restored globally |
| 19 July | Fable 5 plan promotion ends |
| 1 September | Fable 5.1 and Mythos 5.1 launch |
| 22 September | Opus 5.5 launches, beating Fable 5.1 on Anthropic’s table |
Anthropic suspended access because it could not verify nationality in real time. According to Anthropic’s notice on redeploying Fable 5, it then added a classifier that blocks a bypass technique reported by Amazon researchers in “over 99% of cases” (Anthropic-reported).
What is the difference between Claude Fable and Mythos?
Fable and Mythos are the same underlying model with different safeguards. Anthropic called Mythos 5 “the same underlying model as Fable 5, but with the safeguards lifted in some areas.” Fable runs safety classifiers for every user, while Mythos relaxes them for vetted organisations doing cyberdefence or life-sciences research. Mythos 5 was priced the same as Fable 5.
At the June launch, Mythos 5 went to cyberdefenders and infrastructure providers through Project Glasswing, a group of more than 40 companies including Microsoft, Apple and AWS.
Who can get Claude Mythos 5.1?
Mythos 5.1 is available only through Anthropic’s trusted access programmes, and access is “currently limited to US organizations,” according to Anthropic. There are two routes: the Cyber Verification Program (CVP) for security teams and the Life Sciences Verification Program (LSVP), which runs in partnership with the US government.
Anthropic plans to expand beyond the US in coordination with government. Its Fable page says CVP access to Mythos is coming “in the near future”, so expect an application process.
What changed in Claude Fable 5.1 vs Fable 5?
Fable 5.1 scores higher than Fable 5 on every benchmark in Anthropic’s launch table, cuts the cache-read price by 75% to $0.25 per million tokens, and moves the knowledge cutoff from January to June 2026. Its safeguards also fire less often: Anthropic reports 60% fewer cyber interventions per session and far fewer false alarms on basic biology questions.
| Benchmark (Anthropic-reported) | Fable 5 | Fable 5.1 |
|---|---|---|
| Terminal-Bench-Science 0.1 | 24.7% | 52.6% |
| Terminal-Bench 4.0 | 42.0% | 55.8% |
| GDPval-AA v2 (Elo) | 1723 | 1853 |
| OSWorld 2.0 (partial) | 72.9% | 77.9% |
| Humanity’s Last Exam, no tools | 57.8% | 60.9% |
| AutomationBench | 17.1% | 31.4% |
| CursorBench 3.2.0 | 70.5% | 73.4% |
Terminal-Bench-Science more than doubled, within a stated standard error of 3.5 to 4.5 points. All figures come from the Claude Fable 5.1 and Mythos 5.1 announcement. Anthropic’s later Opus 5.5 table uses newer benchmark versions, so don’t mix the two.
Fable 5.1 now permits vulnerability discovery and defensive security work, while penetration testing, exploit generation and binary scanning are still redirected. Bio safeguards fire “85% less often” on benign elementary biology and medical questions (Anthropic-reported).
How much does Claude Fable cost?
Claude Fable 5.1 costs $10 per million input tokens and $50 per million output tokens on the Claude API. Cache reads cost $0.25, 5-minute cache writes $12.50 and 1-hour cache writes $20, and the Batch API is 50% off. It has a 1M-token context window and a maximum output of 128K tokens per response.
| Spec | Claude Fable 5 | Claude Fable 5.1 |
|---|---|---|
| API model ID | claude-fable-5 | claude-fable-5-1 |
| Input / output (per 1M tokens) | $10 / $50 | $10 / $50 |
| Cache write, 5-minute / 1-hour | $12.50 / $20 | $12.50 / $20 |
| Cache read (per 1M tokens) | $1 | $0.25 |
| Context window | 1M tokens | 1M tokens |
| Max output | 128K tokens | 128K tokens |
| Knowledge cutoff | January 2026 | June 2026 |
| Retirement | Legacy, not before 9 June 2027 | Not before 1 September 2027 |
Figures come from Anthropic’s Claude Fable 5.1 model documentation. US-only inference costs 1.1 times the standard price. Thinking is always on, with a default effort of high.
Anthropic says the cache cut makes Fable 5.1 about 25% cheaper on typical workloads and up to 45% cheaper on highly agentic ones (Anthropic-reported). Artificial Analysis’s Fable 5.1 evaluation found the opposite on its test suite: at max effort, Fable 5.1 cost $3.76 per Intelligence Index task against $3.14 for Fable 5, about 20% more, because it wrote roughly 1.7 times as many output tokens.
Try Xhigh Before Paying For Max
At the Fable 5.1 launch, Artificial Analysis scored xhigh effort just one index point below max, at $2.72 per task instead of $3.76. Set effort explicitly on every API call, cap routine agent runs at xhigh, and reserve max for tasks that already failed lower down.
Is Claude Fable included in Claude Pro?
No. As of 24 September 2026, Claude Fable is not included in Pro usage limits. Pro, Team Standard and seat-based Enterprise Standard seats can use Fable 5 and Fable 5.1 only with pay-as-you-go usage credits. Max plans, Team Premium seats and seat-based Enterprise Premium seats can spend up to 50% of their weekly limits on Fable at no extra cost.
For those tiers, Anthropic’s support article on Fable plan access says “Fable 5 and Fable 5.1 are included as a standard part of your plan.” Usage-based Enterprise and API customers pay API rates. In Claude Code, Fable 5.1 needs v2.1.255 or later.
Read more: Claude Pricing 2026
Anthropic’s Fable product page lists Pro among the plans with access, which is easy to misread: on Pro, every Fable request is paid for on top of your subscription. No Anthropic page documents Fable on the Free plan, so assume it is not available. For no-cost options, see our roundup of free AI tools.
What happened to the Fable 5 promotion?
The Fable 5 promotion ended on 19 July 2026 at 11:59:59 PM PT and never covered Fable 5.1. In Anthropic’s words, it “applied to Fable 5 only. Claude Fable 5.1 was never part of it.” Qualifying Pro and Team Standard seats got a one-time credit when Fable 5 moved to usage credits, but there is no equivalent credit for 5.1.
Why does Claude Fable switch to Opus 4.8 or Opus 5?
Claude Fable runs classifiers that detect higher-risk requests and hand them to an older model. In Fable 5.1, flagged cybersecurity requests go to Claude Opus 4.8 and flagged biology requests go to Claude Opus 5. Rerouted requests are not charged at Fable pricing. At the Fable 5 launch, Anthropic said more than 95% of sessions involved no fallback at all.
On the API, a refusal returns HTTP 200 with a stop_reason of “refusal”, and a request refused before any output is not billed. API customers must configure Anthropic’s Fallback API to get automatic rerouting.
If you build agents on Fable, our guide to AI tools for developers covers alternatives with fewer restrictions.
Set Fallbacks Server-Side
Pass the fallbacks parameter on the API instead of catching refusals in your own code. Anthropic’s fallback credit refunds the prompt-cache cost of switching models, so a long cached context is not billed twice when a classifier fires mid-session. Log every stop_reason to see how often your workload trips the safeguards.
Does Anthropic keep your Claude Fable data for 30 days?
By default, yes. Anthropic keeps Fable traffic for 30 days for safety monitoring and says it is not used for training. At the Fable 5 launch this retention was mandatory for all Mythos-class traffic. As of Fable 5.1, zero data retention is offered to eligible customers as an interim option until Enterprise Frontier Safeguards are ready.
Enterprise Frontier Safeguards will keep monitoring data in the customer’s own cloud with customer-managed keys, rolling out from fall 2026 at no separate charge. Teams handling learner records should confirm zero-retention eligibility first.
What do reviewers and users say about Claude Fable?
Reviewers agree Fable was a big capability jump for long, autonomous work, and they complain just as consistently about its cost, token hunger and strict safeguards. Pro subscribers also objected to the usage-credit requirement and the shifting plan deadlines between June and July 2026. Few doubt its ability; many doubt its value for money.
In Simon Willison’s review of Fable 5, he called it “relentlessly proactive” and “something of a beast. It’s slow, expensive.” He warned it “will quite happily burn $12 in tokens inventing new ways to debug your CSS.”
Zvi Mowshowitz’s Fable 5 capabilities review said “The jump is impossible to miss,” and relayed Andrej Karpathy calling it a “major-version-bump-deserving step change forward.” He found Fable “is often net cheaper” than Opus despite higher token prices, but warned “The classifiers are not messing around.”
Is Claude Fable still worth it now that Opus 5.5 is out?
For most work, no. Claude Opus 5.5, launched on 22 September 2026, beats Fable 5.1 on every row of Anthropic’s own benchmark table and scores 58 to Fable’s 53 on the Artificial Analysis Intelligence Index. At $4 input and $20 output per million tokens, it costs 40% of Fable’s price, so Fable is 2.5 times as expensive per token.
Artificial Analysis has since re-based its index, which is why Fable 5.1’s launch score of 66 now reads 53. Fable keeps narrow edges: it used about 78,000 output tokens per task against 119,000 for Opus 5.5, and some reviewers still prefer it for the hardest coding problems.
Our Claude Opus 5.5 vs Fable 5.1 comparison covers when the premium pays off. If price matters most, our GPT-6 Sol vs Claude Opus 5.5 comparison looks at OpenAI’s cheaper option.
Is there a Claude Fable 5.5?
No. As of 24 September 2026, Anthropic has not announced a Fable 5.5. Its Opus 5.5 announcement names only Claude Sonnet 5.5 and Haiku 5.5 as arriving “in the coming weeks,” with no dates. Fable 5.1 remains the current Fable model, and Anthropic’s docs say it will not be retired before 1 September 2027.
Conclusion
Claude Fable brought Mythos-class capability to the public, but it is now a specialist tool rather than a default. It costs 2.5 times as much as Opus 5.5 per token, sits behind usage credits on Pro, and routes flagged cyber and biology work to older models.
Your next step: on Max or a Premium seat, keep Opus 5.5 as your default and send Fable 5.1 only the tasks Opus has already failed. On Pro or the API, run your hardest recurring tasks on both models and compare cost per completed task before buying credits. US cyberdefence and life-sciences teams can apply to the CVP or LSVP for Mythos 5.1.
FAQ
Q1. Do I need usage credits to use Claude Fable 5.1?
It depends on your plan. On Pro, Team Standard and seat-based Enterprise Standard seats, yes: Fable 5 and Fable 5.1 sit outside your usage limits and run on pay-as-you-go credits. Max plans and Premium seats can use Fable within 50% of their weekly limits at no extra cost. Fable 5.1 was never part of any promotional allowance.
Q2. What is the difference between Fable 5.1 and Mythos 5.1?
They are the same model with different safeguards. Mythos 5.1 has more permissive safeguards and is available only to vetted organisations in Anthropic’s Cyber Verification Program or Life Sciences Verification Program, currently limited to US organisations. Fable 5.1 is open to everyone but sends flagged cybersecurity requests to Opus 4.8 and flagged biology requests to Opus 5.
Q3. Why does Claude Fable refuse legitimate technical work?
Fable’s classifiers flag requests that look like higher-risk cybersecurity or biology work, even when the purpose is benign. Fable 5.1 intervenes far less often than Fable 5 and now allows vulnerability discovery and defensive security work, but penetration testing, exploit generation and binary scanning are still redirected to Opus 4.8. Vetted US organisations can apply for Mythos access instead.
Q4. Is Claude Fable 5 still available?
Yes. Fable 5 is now a legacy model, still offered through the API as claude-fable-5, and Anthropic says it will not be retired before 9 June 2027. It costs the same $10 input and $50 output per million tokens as Fable 5.1, but cache reads cost $1 instead of $0.25, so most users are better off on 5.1.
Q5. Which model does Claude Code use by default?
As of 24 September 2026, Claude Opus 5.5 is the default model in Claude Code for Pro, Max, Team, Enterprise and Anthropic API users. You can still choose Fable 5.1 if your plan allows it, but you need Claude Code v2.1.255 or later; Fable 5 needs v2.1.170 or later. On Pro, Fable usage is paid for with usage credits.
Q6. Does Fable 5.1 use more of my plan?
On Max plans and Team or Enterprise Premium seats, Fable can use up to 50% of your weekly limits at no extra cost. It is also a heavy model: it costs 2.5 times as much per token as Opus 5.5, and Artificial Analysis found Fable 5.1 writes about 1.7 times as many output tokens per task as Fable 5.
Q7. Is Opus 5.5 better than Fable 5.1?
On published tests, yes. Opus 5.5 beats Fable 5.1 on every row of Anthropic’s own benchmark table and scores 58 against 53 on the Artificial Analysis Intelligence Index, at 40% of the per-token price. Fable 5.1 uses fewer output tokens per task, and some reviewers still prefer it for the very hardest coding problems.
