DataLLM LabClaude Fable 5: Anthropic's #1 Model, Banned and Back
Claude Fable 5 is the most dramatic model story of 2026: Anthropic's top-tier frontier model launched at #1 on the independent Artificial Analysis Index, then vanished worldwide about three days later under a US government export-control order - and came back on July 1. This is what Fable 5 is, exactly what happened during the ~18-day blackout, where it stands now (restored, but conditionally), and the availability lesson every team building on a single closed model should take from it.
What Claude Fable 5 is
Claude Fable 5 is Anthropic's top-tier "Mythos-class" general frontier model — a step above Claude Opus 4.8. It launched on June 9, 2026 straight to #1 on the independent Artificial Analysis Intelligence Index at 64.9, roughly 5 points ahead of the nearest non-Anthropic model, and is priced at a premium $10 / $50 per 1M input/output tokens (about double Opus 4.8's input price). For public use, Fable 5 is the same underlying model as Claude Mythos 5 with extra guardrails for high-risk queries and an Opus 4.8 fallback that catches safety-flagged prompts.
The 18-day blackout: timeline
Fable 5 went from launch to fully dark worldwide in about three days, and stayed down for roughly 18:
| Date (2026) | What happened |
|---|---|
| Jun 9 | Fable 5 launches — #1 on the Artificial Analysis Index (64.9), $10/$50 |
| Jun 12 | US export-control directive bars all foreign nationals → Anthropic suspends Fable 5 & Mythos 5 worldwide to comply |
| Jun 13 | Anthropic calls it a "misunderstanding"; review finds only simple, known vulnerabilities. No restoration timeline. |
| Jun 26 | Government approves Mythos 5 for a limited set of US (Project Glasswing) partners |
| Jun 30 | US Commerce Department lifts the export controls — conditionally |
| Jul 1 | Fable 5 restored globally, rolling out across API, Claude.ai, Claude Code & Cowork (50% of weekly limits through Jul 7) |
The trigger was a government-cited "jailbreak" of a Fable 5 cybersecurity safeguard — reportedly originating from an Amazon research report — that got a prompt to identify software vulnerabilities. Anthropic characterized it as a narrow technique that surfaced only vulnerabilities other public models could already find.
Where it stands now
As of July 1, 2026, the ~18-day worldwide blackout is over — but the two models are in different states:
- Fable 5 — restored globally. Rolling out from July 1 across the Claude Platform (API), Claude.ai, Claude Code, and Claude Cowork; cloud providers (AWS, Google Cloud, Microsoft Foundry) are re-enabling it with no firm per-cloud date. On Pro/Max/Team/select Enterprise plans it's included up to 50% of weekly usage limits through July 7, then via usage credits.
- Mythos 5 — partially restored. Still limited to a set of approved US organizations (Project Glasswing) since June 26; broader domestic and international access is pending with no GA timeline.
- The lift is conditional and revocable. The clearance came after Anthropic agreed to proactively detect security risks, give US agencies early/pre-release model access, report malicious activity, and work toward common industry security standards — and it deployed a new classifier that blocks the reported technique in >99% of cases, with the Opus 4.8 fallback. The government reserves the right to reimpose restrictions.
Fable 5 vs Mythos 5
The two names cause confusion, so to be precise: Mythos 5 is the base model; Fable 5 is the safety-wrapped public configuration of it. Fable 5 adds guardrails for cyber/bio/chem/distillation-style queries and the Opus 4.8 fallback (its full config name is "Claude Fable 5 (Adaptive Reasoning, Max Effort, Opus 4.8 Fallback)"). Mythos 5 runs with fewer safeguards and is only offered to vetted partners — which is why, post-episode, Fable 5 is back for everyone while Mythos 5 stays gated.
The availability lesson
Strip away the drama and Fable 5 is a live case study in single-vendor concentration risk. A model can launch at #1 and be fully unavailable worldwide within ~90 minutes of a government order, stay down for weeks with no timeline, and even its return can be conditional and revocable. That's an availability and compliance risk, not a quality one — and it doesn't show up in any benchmark.
The practical hedge showed itself during the blackout: open-weights models were completely unaffected (the directive targeted only Anthropic's two closed models), so teams pivoted to DeepSeek, GLM, Qwen, and Kimi and kept shipping. Tellingly, Anthropic's own fix routes flagged Fable 5 queries to a fallback model — the same pattern you should run at the product level: multi-provider routing with a warm open-weights fallback, so one model going dark never takes your app down with it. (For alternatives, see our Fable 5 alternatives guide; for the broader policy picture, AI export controls & open weights.)
Never let one model take your product down
Route Claude, GPT, Gemini and open-weights models (GLM, DeepSeek, Qwen) through one OpenAI-compatible endpoint with automatic failover — a warm fallback for exactly this.
FAQ
What is Claude Fable 5?
Anthropic's top "Mythos-class" frontier model, above Opus 4.8. Launched Jun 9, 2026 at #1 on the Artificial Analysis Index (64.9), $10/$50 per 1M. Same base as Mythos 5 + extra guardrails + an Opus 4.8 fallback.
Is Claude Fable 5 available now?
Yes — restored globally from July 1, 2026 after the US lifted export controls on June 30. Rolling out across API/Claude.ai/Claude Code/Cowork; 50% of weekly limits through July 7, then credits. Rollout is phased and revocable.
Why was it suspended?
A June 12, 2026 US export-control directive barred all foreign nationals from the two models; with no way to verify nationality in real time, Anthropic disabled them worldwide. Trigger: a government-cited jailbreak (reportedly from an Amazon report) that Anthropic called a misunderstanding.
How much does Fable 5 cost?
$10 / $50 per 1M input/output tokens — about 2x Opus 4.8 on input. A premium top-tier model.
Fable 5 vs Mythos 5?
Same underlying model. Fable 5 = public config with extra guardrails + Opus 4.8 fallback (restored globally). Mythos 5 = fewer safeguards, gated to approved US partners (still limited).
Is Fable 5 better than Opus 4.8 or GPT-5?
It launched #1 on the independent Artificial Analysis Index at 64.9, above Opus 4.8 and ~5 pts ahead of the nearest non-Anthropic model. We can't benchmark it first-party (it was suspended / is phased), so we cite Artificial Analysis.
Could it be banned again?
Yes — the lift is explicitly revocable; the government can reimpose restrictions if circumstances change or Anthropic misses its commitments. Availability is now a real design factor.
What if I need guaranteed availability?
Run multi-provider routing with a warm open-weights fallback (DeepSeek/GLM/Qwen/Kimi were unaffected). A gateway like DataLLM Lab keeps a fallback ready so one model's outage doesn't take your product down.