★ Reading this for free? Get 20 structured AI courses + per-chapter AI tutor — the first chapter of every course free, no card.Start free in 30 seconds

Anthropic · Closed-API Model · Suspended

Claude Fable 5 & Mythos 5: The #1 Model, Switched Off Worldwide

Anthropic released Claude Fable 5 on June 9, 2026 — its first publicly available "Mythos-class" model, a tier above Opus — and within days it sat at #1 on the Artificial Analysis Intelligence Index, #1 on LMArena, and on top of SWE-bench Pro. Three days later, on June 12, 2026, a US government export-control directive forced Anthropic to suspend Fable 5 and Mythos 5 globally for every user, with no notice. As of mid-June it is still offline. This page is not a buy recommendation — Fable 5 is unavailable — it is the clearest case study to date for why a model you can run locally is one nobody can take away.

📅 Published: June 19, 2026🔄 Last Updated: June 19, 2026✓ Manually Reviewed

Status: suspended. Claude Fable 5 and Mythos 5 are unavailable worldwide as of mid-June 2026 following a US export-control order. Do not design a workflow around them. For models that cannot be remotely revoked, see DeepSeek V4 and GLM-5; for a still-online Anthropic option, see Claude Opus 4.7.

Key takeaways

  • Released June 9, 2026 — Anthropic's first public Mythos-class model, a rung above Opus.
  • #1 everywhere — top of the Artificial Analysis Index (~64.9), LMArena (~1510 Elo), and SWE-bench Pro (~80.3%).
  • Fable 5 = safeguarded public version; Mythos 5 = unrestricted, Project-Glasswing-only. Same model underneath.
  • Suspended worldwide June 12, 2026 by US export-control order over offensive-cyber risk — still offline.
  • The lesson: a cloud model can be switched off by a vendor or a government overnight. An open-weight model on your hardware cannot.

What happened, in one paragraph

On June 9, 2026 Anthropic put its most capable model in front of the public for the first time. Claude Fable 5 was the general-availability face of a new internal tier Anthropic calls "Mythos-class" — explicitly positioned above the Opus line. It launched on the Claude API, Claude.ai, Amazon Bedrock, Vertex AI and Microsoft Foundry, and it immediately topped the public leaderboards. TechCrunch noted the irony that Anthropic shipped its most powerful public model days after warning that AI was getting too dangerous.

On June 12, 2026 the US government delivered an export-control directive ordering the immediate suspension of Fable 5 and Mythos 5 for any foreign national — anywhere. Anthropic could not gate access by nationality in real time, so it took both models offline for the entire planet. The product that was #1 on every index 72 hours earlier was simply gone, and as of mid-June it has not come back.

Specs at a glance

VendorAnthropic
TierMythos-class (above Opus)
Release dateJune 9, 2026
Current statusSuspended worldwide (since June 12, 2026)
VariantsFable 5 (safeguarded, GA) · Mythos 5 (unrestricted, Project Glasswing)
Pricing (when live)$10 input / $50 output per 1M tokens (both variants)
LicenseProprietary (Anthropic Terms of Service)
Local self-hostable?No
Can be remotely revoked?Yes — and was
Suspension causeUS export-control directive (offensive-cyber capability)

Specs from Anthropic's launch materials and Claude API docs, June 2026. The model is offline, so figures cannot currently be re-verified against a live endpoint.

Fable 5 vs Mythos 5: same model, different leash

The two names describe one Mythos-class model with two access policies.

Claude Fable 5 — general availability, with safety classifiers

The public version. It ships with safeguards that decline requests in sensitive domains — primarily cybersecurity and biology. Anthropic says it tuned these conservatively: they occasionally catch benign requests but fire in under 5% of sessions on average. This is the model that briefly topped the leaderboards before the suspension.

Claude Mythos 5 — restricted, no classifiers (Project Glasswing)

The unrestricted version with the safety classifiers removed, released only through Project Glasswing to a small set of pre-approved partners — US government cyber defenders, with select biomedical research organizations slated next. Same raw capability as Fable 5, without the public-facing guardrails. It was suspended alongside Fable 5 on June 12.

Benchmarks, as reported at launch

For the three days it was live, Fable 5 led the public boards. These are the numbers Anthropic and the trackers reported on launch; we present them as reported, since the model is now offline and cannot be independently re-run.

MetricClaude Fable 5Claude Opus 4.8GPT-5.5Gemini 3.1 Pro
SWE-bench Pro (agentic coding)~80.3%~69.2%~58.6%~54.2%
Artificial Analysis Index (composite)~64.9 (#1)
LMArena text (Elo)~1510 (#1)

Sources: Anthropic launch announcement (June 9, 2026) and Claude API docs · Artificial Analysis Intelligence Index (reported ~64.9, #1, ~5 pts clear) · LMArena text leaderboard (reported ~1510 Elo, #1) · SWE-bench Pro launch figures. Comparator columns shown only where a like-for-like figure was reported. Anthropic announcement · Artificial Analysis

The suspension, in detail

Per reporting from CNBC, Al Jazeera, Tom's Hardware and others, the US government issued an export-control directive on June 12, 2026 invoking national-security authorities. It named Fable 5 and Mythos 5 and ordered access suspended for any foreign national — inside or outside the United States, up to and including Anthropic's own foreign-national employees.

The stated concern was offensive cybersecurity. Reports describe Fable 5 demonstrating a near-"superhuman" ability to read an arbitrary codebase and surface previously unknown vulnerabilities — exactly the capability a hostile actor would want for finding exploits at scale. Because Anthropic has no way to verify a user's nationality in real time at the API layer, the only compliant move was to disable both models for every customer on Earth. Security commentators (Snyk, Simon Willison) have flagged this as what looks like the first government-forced takedown of a deployed frontier model.

Note: there was no notice, no grace period, no per-customer opt-out, and no refund mechanism described. Teams that had wired Fable 5 into a product on June 9 had a dead endpoint by June 12. Anthropic has said it is working to restore access "as soon as possible," with no committed date.

The local-vs-cloud lesson

This is the cleanest illustration yet of a risk that is easy to wave away until it bites: a cloud model is access, not ownership. Your ability to call Fable 5 sat at the intersection of two things you do not control — Anthropic's business decisions and a regulator's reach. The moment the second one moved, your access evaporated, and no amount of being a paying customer changed that.

An open-weight model is the opposite shape of dependency. Once the weights for DeepSeek V4 or GLM-5 are on your own disk, there is no endpoint anyone can disable, no directive that can reach into your rack, and no Terms-of-Service change that can deprecate the model out from under you. It will run on the same hardware in a year regardless of what any vendor or government decides. That is not an argument that local models are better — at the frontier they trail — it is an argument that they are durable, and durability is a feature you only notice the day the cloud one disappears.

The pragmatic posture most teams land on: use the strongest available managed model for the hard cases, but keep a self-hosted open-weight model as a fallback that cannot be revoked. Fable 5 is the reason that fallback is not paranoia.

What to use instead

Fable 5 is unavailable, so there is nothing to migrate to it. Two reasonable directions:

OptionTypeWhy
DeepSeek V4Open-weight (MIT)Cannot be remotely revoked. Closest frontier-class open-weight general model — the durable fallback.
GLM-5Open-weight (MIT)Strong open-weight reasoning/coding, smaller hardware footprint. Self-host and you own access.
Claude Opus 4.7Closed API (Anthropic)Still online. The flagship Anthropic line — fine if you want a managed model and accept the cloud-dependency risk.
Gemini 3.1 ProClosed API (Google)Available frontier alternative with 1M context — same structural revocation risk as any cloud model.

If your reason for wanting Fable 5 was raw capability, Opus 4.7/4.8 and GPT-5.5 are the strongest things you can actually call today. If your reason was that you cannot afford to have a model vanish on you, the answer is local — full stop.

Frequently asked questions

Is Claude Fable 5 available right now?
No. As of mid-June 2026, Claude Fable 5 and Claude Mythos 5 are suspended worldwide. Anthropic disabled both models on June 12, 2026 to comply with a US government export-control directive that barred access by any foreign national. Because Anthropic cannot filter users by nationality in real time, the only way to comply was to switch the models off for everyone — including US customers and Anthropic's own foreign-national staff. Anthropic has said only that it is working to restore access "as soon as possible"; no return date has been announced. Do not build a workflow on Fable 5 right now — there is nothing to call.
Why was Claude Fable 5 suspended?
The US government issued an export-control directive on June 12, 2026 citing national-security authorities. The stated concern was Fable 5's offensive-cybersecurity capability — reports describe the model showing a "superhuman" ability to read a codebase and surface previously unknown software vulnerabilities, which a foreign adversary could use to find exploits at scale. The directive named Fable 5 and Mythos 5 specifically and suspended access for any foreign national inside or outside the US. This appears to be the first time a government has forced a publicly deployed frontier AI model offline.
What is the difference between Claude Fable 5 and Claude Mythos 5?
They are the same underlying Mythos-class model. Claude Fable 5 is the general-availability version with safety classifiers that decline requests in sensitive areas (primarily cybersecurity and biology); Anthropic tuned those conservatively and says they fire in under 5% of sessions. Claude Mythos 5 is the unrestricted version with no classifiers, released only through Project Glasswing to a small set of pre-approved partners — US government cyber defenders, and (per Anthropic) select biomedical research organizations next. Mythos was the internal, restricted tier; Fable was the safeguarded public face of it. The June 12 directive suspended both.
How much did Claude Fable 5 cost?
Fable 5 and Mythos 5 were both priced at $10 per million input tokens and $50 per million output tokens — roughly 2× Claude Opus 4.8 ($5 / $25). Through June 22, 2026 Fable 5 was also bundled at no extra cost into Pro, Max, Team and seat-based Enterprise plans on Claude.ai, with metered API pricing to follow. By comparison, the earlier Glasswing-only access ran at $25 / $125 per million tokens. Pricing is academic at the moment: the model is offline, so you cannot be billed for it.
How good was Claude Fable 5 on benchmarks?
It topped essentially every public leaderboard for the three days it was live. Reported figures: ~80.3% on SWE-bench Pro (the harder agentic coding test, vs ~69% for Opus 4.8 and far ahead of GPT-5.5 and Gemini 3.1 Pro), #1 on the Artificial Analysis Intelligence Index with a composite ~64.9 (about 5 points clear of any other lab's best), and the #1 spot on the LMArena text leaderboard near 1510 Elo. Treat these as reported launch numbers; they will not be independently re-confirmed while the model is suspended.
What can I use instead of Claude Fable 5?
For something that cannot be remotely revoked, use an open-weight model you host yourself. DeepSeek V4 (MIT-licensed) and GLM-5 are the closest frontier-class open-weight options — once the weights are on your hardware, no vendor or government directive can switch them off. If you want a managed Anthropic model that is still available, Claude Opus 4.8 / 4.7 remain online ($5 / $25 and up). But Fable 5's suspension is exactly the case for keeping a local fallback: a cloud model is a dependency you do not control.
Could a US export order really take away a model I am paying for?
It already did. The June 12, 2026 directive applied to a live, paid, publicly available product and pulled it for everyone with no notice and no opt-out. There was no refund mechanism, no grace period, and no way for an individual customer to keep access. That is the structural risk of any API-only model: your access exists at the intersection of the vendor's business decisions and the regulator's reach, and neither is under your control. A model whose weights live on your own disk has no such off switch.

Build on a model nobody can switch off

Fable 5 proved the point: a cloud model can be gone in 72 hours. The Local AI Master deployment course shows you how to run open-weight models like DeepSeek V4 and GLM-5 on your own hardware — so your stack keeps working no matter what a vendor or a government does next.

See the deployment course →

Related models

🎯
AI Learning Path

Go from reading about AI to building with AI

20 structured courses. Hands-on projects. Runs on your machine. Start free.

Or own it for life — Lifetime $149 $599, pay once
LM

Written by the Local AI Master Team

The team behind Local AI Master

We build Local AI Master around practical, testable local AI workflows: model selection, hardware planning, RAG systems, agents, and MLOps. The goal is to turn scattered tutorials into a structured learning path you can follow on your own hardware.

✓ Local AI Curriculum✓ Hands-On Projects✓ Open Source Contributor
More on AI Models Directory
See the full AI Models Directory guide.
📚
Free · no account required

Grab the AI Starter Kit — career roadmap, cheat sheet, setup guide

No spam. Unsubscribe with one click.

🎯
AI Learning Path

Found your model? Now build something with it.

20 hands-on courses — RAG, agents, fine-tuning — all running locally. First chapter free, no card.

Or own it for life — Lifetime $149 $599, pay once
Free Tools & Calculators