Claude Fable 5 Is Here: Mythos-Class Power, With Strings Attached
Last updated: June 2026
On June 9, 2026, Anthropic released Claude Fable 5, the first Mythos-class model available to the general public, alongside a restricted upgrade called Claude Mythos 5. Anthropic describes Fable 5 as the most capable model it has ever made generally available, with its largest leads showing up on long, complex tasks rather than quick one-off queries.
That headline is real. But two details most of today's coverage is glossing over matter more for how you actually use the model: on a standard subscription you can only access Fable 5 through June 22, and every conversation you have with it is now retained for 30 days. We will cover the full specs, then the parts that affect your wallet and your data.
- Fable 5 is a Mythos-class model made safe for public use. A set of classifiers routes cybersecurity, biology, chemistry, and model-distillation queries to Claude Opus 4.8 instead; Anthropic says this fallback affects fewer than 5% of sessions.
- It is included on Pro, Max, Team, and seat-based Enterprise plans only through June 22, 2026. From June 23, using it requires usage credits (effectively pay-per-use) until Anthropic adds capacity.
- All traffic on Mythos-class models is now subject to mandatory 30-day data retention, including customers who previously held zero-retention agreements. Anthropic says the data is used only for safety, never for training.
What Are Claude Fable 5 and Mythos 5?
"Mythos" is Anthropic's name for a tier of models that sits above its Opus line in capability. The first, Claude Mythos Preview, launched in April 2026 through Project Glasswing and was never released to the public. It was handed only to a small group of cybersecurity defenders and critical-infrastructure providers because of how effectively it could find and exploit software vulnerabilities.
Fable 5 and Mythos 5 are the same underlying model. The only difference is safeguards. Mythos 5 has its cybersecurity guardrails lifted and remains restricted to approved Glasswing partners, with a small group of biology researchers to follow. Fable 5 is the public version, shipped with classifiers that block certain high-risk requests. Anthropic notes that both names come from words meaning "that which is told" (the Latin fabula and Greek mythos) and that the safeguards are the entire distinction between them.
When Fable 5's classifiers flag a request related to cybersecurity, biology, chemistry, or model distillation, the answer is handled by Claude Opus 4.8 instead, and you are told when it happens. Anthropic says fewer than 5% of sessions trigger this fallback, and because Opus 4.8 is itself a strong model, a fallback is a far better experience than a flat refusal.
Claude Fable 5 Benchmarks
The table below shows Fable 5 against the other leading models you can actually access today: Opus 4.8, GPT-5.5, and Gemini 3.1 Pro. Where Anthropic reported separate Fable 5 and Mythos 5 figures within a few points of each other, the higher of the two is shown.
| Benchmark | Fable 5 / Mythos 5 | Opus 4.8 | GPT-5.5 | Gemini 3.1 Pro |
|---|---|---|---|---|
| Agentic coding (SWE-Bench Pro) | 80.3% | 69.2% | 58.6% | 54.2% |
| Agentic coding (Terminal-Bench 2.1) | 88.0%* | 82.7% | 83.4% | 70.7% |
| Knowledge work (GDPval-AA) | 1932 | 1890 | 1769 | 1314 |
| Knowledge work, vision (GDPval) | 29.8% | 22.5% | 24.9% | 16.7% |
| Spatial reasoning (Blueprint-Bench 2) | 38.6% | 14.5% | 36.2% | 26.5% |
| Tool use (AutomationBench) | 17.4% | 15.5% | 12.9% | 9.6% |
| Multidisciplinary reasoning (Humanity's Last Exam, with tools) | 64.5%* | 57.9% | 52.2% | 51.4% |
| Cybersecurity (ExploitBench, Cap%) | 78.0%* | 40.0% | 34.0% | — |
* On starred benchmarks, Fable 5's safeguards route cybersecurity- and biology-related questions to Opus 4.8, so the public Fable model performs closer to Opus 4.8's score than the figure shown. The starred numbers reflect the higher Mythos 5 result. Full methodology is in Anthropic's system card.
Software engineering. This is Fable 5's strongest area. During early testing, Stripe reported that the model performed a codebase-wide migration across a 50-million-line Ruby codebase in a single day, work the company estimated would have taken a team more than two months by hand. On Cognition's FrontierCode evaluation, which grades whether a model can pass difficult coding tasks while meeting production-codebase standards, Fable 5 scored highest among frontier models even at medium effort.
Vision. Fable 5 is the new state of the art on vision tasks. It can pull precise figures from detailed scientific charts and rebuild a web app's source code from screenshots alone. As a demonstration, it finished the Game Boy Advance title Pokemon FireRed using only raw screenshots, with no maps or navigation aids, something earlier Claude models could not manage even with purpose-built helper tools.
Memory and long context. The model holds focus across millions of tokens and improves its own work using notes it keeps as it goes. When Anthropic had it play the deck-builder Slay the Spire, giving it persistent file-based memory improved its performance three times more than the same memory helped Opus 4.8.
Knowledge work. Analytics firm Hex reported that Fable 5 was the first model to clear 90% on its core benchmark of complex, long-running analytical tasks, roughly a 10-point jump over Opus 4.8. Anthropic also showed the model autonomously playing the factory-building game Factorio and building a working solar-system simulation that derives orbital motion from first principles to predict eclipses.
Pricing and Access: The June 22 Cliff
Through the Claude API, Fable 5 costs $10 per million input tokens and $50 per million output tokens, roughly double the price of Opus 4.8.
| Pricing (per million tokens) | Claude Fable 5 | Claude Opus 4.8 |
|---|---|---|
| Input | $10 | $5 |
| Output | $50 | $25 |
Access through subscriptions is where pro-consumer readers should pay attention, because Anthropic is rolling it out in stages and expects demand to be hard to predict:
- From launch through June 22, 2026, Fable 5 is included on Pro, Max, Team, and seat-based Enterprise plans at no extra cost.
- On June 23, Anthropic removes Fable 5 from those plans. Using it after that requires usage credits, which is effectively pay-per-use billing, similar to the API.
- Anthropic says it aims to restore Fable 5 as a standard plan feature "as quickly as we can" once capacity allows, and may extend the included window if it can, but it has not committed to a date.
In plain terms: if you are on a Max plan, your included access to Fable 5 is currently scheduled to run only through June 22. After that you are paying per usage until Anthropic adds enough capacity to fold it back into subscriptions. That is worth knowing before you build a workflow around it.
One more consumption note from early users: several report that Fable 5 burns through plan usage at roughly twice the rate of Opus 4.8 per task. So even inside the included window, heavy users will hit their limits faster than they are used to. Treat the next two weeks as a window to test the model, not to depend on it.
The Privacy Shift: Mandatory 30-Day Data Retention
Alongside the launch, Anthropic changed how it handles data for Mythos-class models. It now requires 30-day retention for all traffic on Fable 5, Mythos 5, and future models at similar capability levels, on both Anthropic's own surfaces and third-party platforms. Notably, this applies even to business customers who previously held zero-retention agreements.
Anthropic's stated reason is safety. The company says the retained data will be used only to defend against sophisticated attacks, including novel jailbreaks and attacks that play out across many separate requests, and to reduce the false positives its conservative safeguards currently produce. It says the data will not be used to train new models or for any non-safety purpose, and that it has added protections including logging every instance of human access and deleting the data after 30 days in almost all cases.
For a site built on local-first and data-sovereignty principles, this is the trade-off stated out loud. Fable 5 is a model you rent, not one you own, and that rental now comes with non-negotiable logging attached. This is not cause for alarm, since the rationale is legitimate and the protections are real, but it is cause to know exactly what you are agreeing to before you paste proprietary code, client documents, or sensitive personal information into the model.
If that arrangement does not sit right with you, the alternative is not a different cloud provider. It is owning the stack yourself.
Closed Power vs. Open Control
Fable 5 is a closed model that runs only on Anthropic's servers. It cannot be downloaded, and it cannot run offline or on hardware you control. That is the core constraint behind both the access cliff and the retention policy: when the model lives on someone else's infrastructure, they set the terms, and the terms can change on launch day.
The open-weight world is the counterweight. Models such as Google's Gemma 4, Alibaba's Qwen, DeepSeek, and GLM-5.1 can be downloaded and run on your own machine, with no per-query fees, no usage cliff, and no retention policy, because nothing leaves your network.
We will be straight about the gap: open-weight models do not match Fable 5 on the hardest long-horizon agentic tasks today. Digital sovereignty still carries a real capability cost at the frontier. But for a large share of everyday work, including drafting, summarizing, chatting with your own documents, and routine coding help, a capable open model on hardware you own is already enough, and it will never change its terms on you mid-project. If you want to go that route, our guide to the best mini PCs for local AI is the place to start.
Safeguards and Why the Limits Exist
The reason Fable 5 ships with such broad guardrails comes down to what Mythos-class models can do. As we covered when Mythos Preview first surfaced, these models are state-of-the-art at discovering and exploiting software vulnerabilities, and they are strong at agentic hacking, chaining together reconnaissance, discovery, and lateral movement rather than just finding a single flaw. The same skill that helps a defender harden a system helps an attacker break into one. That dual-use problem is why the safeguards exist.
Anthropic says it red-teamed the classifiers hard. An external bug bounty ran more than 1,000 hours without producing a universal jailbreak, though the company notes the UK AI Security Institute made progress toward one in a brief testing window. In its blocking mode, Fable 5 made no progress on offensive-cyber evaluation tasks. On the alignment side, Anthropic's automated assessment found Mythos 5's rate of misaligned behavior to be low and similar to Opus 4.8, and since Fable 5 is the same underlying model, its alignment profile should match.
For home-network readers, the practical takeaway is unchanged from our earlier Mythos coverage. The public Fable 5 model cannot be used for offensive cybersecurity, so it is not a direct threat to your router or your local services. But it is one more marker on a capability curve that keeps climbing, and that curve is exactly why hardening the network you own, with strong admin credentials, current firmware, and network segmentation, remains worth the effort.
Frequently Asked Questions
Is Claude Fable 5 free?
It is included at no extra cost on Pro, Max, Team, and seat-based Enterprise plans, but only through June 22, 2026. From June 23, using Fable 5 requires usage credits (pay-per-use) until Anthropic restores it as a standard plan feature, which it has not yet dated. Through the API, it is billed per token from day one.
What is the difference between Claude Fable 5 and Claude Mythos 5?
They are the same underlying model. Fable 5 is the public version with safeguards that route cybersecurity, biology, chemistry, and distillation queries to Opus 4.8. Mythos 5 has the cybersecurity safeguards lifted and is restricted to approved Project Glasswing partners, with select biology researchers to follow.
Can I run Claude Fable 5 locally?
No. Fable 5 is a closed model that runs only on Anthropic's servers; it cannot be downloaded or run offline. If local operation matters to you, use open-weight models such as Gemma 4, Qwen, or DeepSeek, which you can run on your own hardware.
How much does Claude Fable 5 cost through the API?
$10 per million input tokens and $50 per million output tokens, roughly double the cost of Claude Opus 4.8 ($5 input / $25 output). Anthropic notes this is less than half the price of the earlier Mythos Preview.
Does Anthropic store my Claude Fable 5 conversations?
Yes. Anthropic now requires 30-day data retention for all traffic on Mythos-class models, including customers who previously had zero-retention agreements. The company says the data is used only for safety purposes, never for training, and is deleted after 30 days in almost all cases.
Is Claude Fable 5 better than Opus 4.8?
On benchmarks, yes, particularly on long, complex, and multi-step tasks, where its lead is largest. In day-to-day use the gap is narrower, and because Fable 5 falls back to Opus 4.8 on flagged topics and costs more, Opus 4.8 remains the more practical default for many users.
