Claude Fable 5 Was Gone — the Outage Playbook (Now Restored)
Claude Fable 5 was banned June 12 and restored July 1, 2026. The action checklist developers and subscribers used during the outage — and what to do now that it is back.

Update — July 1, 2026: Fable 5 is back worldwide — restored on July 1 after the US export-control directive was withdrawn on June 30. Through July 7, Pro/Max/Team plans include Fable 5 in up to 50% of the weekly usage cap; after that it moves to usage credits ($10/$50 per million tokens). This post is now a historical playbook of the June 12 – July 1 outage; the fallback patterns below still apply because the restored model's new safety classifiers occasionally route requests to Opus 4.8. Live status: /is-fable-5-back · Full story: Claude Fable 5 is back.
Claude Fable 5 went offline on June 12, 2026 — your codebase, subscription, and API keys did not.
The US export-control ban caught everyone mid-sprint — including teams that had just migrated workflows to Fable 5. This guide was a role-based action list during the 19-day outage: what to do, not a recap of why it happened (ban timeline) or when it would return (restoration story).
Track availability: /status
Developers — what the outage playbook was
1. Switch default model to Opus 4.8 (now optional — Fable 5 is back)
# Before
model = "claude-fable-5"
# Now
model = "claude-opus-4-8" # or "claude-opus-5" since July 24, 2026
Same change in Claude Code, Cursor agent settings, and any CI pipelines that pinned Fable 5.
2. Add a circuit breaker (keep this one)
If claude-fable-5 errors, route to Opus automatically. This advice outlived the outage: post-restoration, the new safety classifiers occasionally reroute routine coding/debugging requests to Opus 4.8 while Anthropic tunes false positives.
3. Log requested vs served model (keep this too)
Fable 5's safeguard fallback already taught this lesson: billing and quality can diverge silently. Keep structured logs.
4. Tune effort, not temperature
Fable 5 removed temperature / top_p. Opus 4.8 still supports them, but Anthropic's direction is the effort parameter on newer models. Match depth to task — most work does not need max effort.
Deep dive: Fable 5 vs Opus 4.8 · vs GPT-5.5 · Prompt design guide
Subscribers — billing and refunds
What changed for paid plans
- Fable 5 could not be selected on any plan from June 12 to July 1 — the June 9–22 included window was frozen while the model was offline.
- After the July 1 restoration, Anthropic reset the billing terms: Fable 5 is included in up to 50% of weekly caps through July 7, 2026, then moves to usage credits at API pricing.
Refunds
During the outage Anthropic did not publish a formal refund program; Forbes (June 14) reported customers asking for refunds after paying for access to a model that lasted three days. With access restored and the included window reset, the refund question has largely resolved itself — but if you still have a billing dispute from the outage window:
- Check Anthropic support for billing policy updates.
- Document your usage window (June 9–12) if disputing charges.
More on post-window billing: after the included window.
API and enterprise users
- Audit API keys — re-point services to
claude-fable-5where you want it back. - Update runbooks — incident docs should name Opus 4.8 as the fallback default (it still is, via the new safety classifiers).
- Review contracts — enterprise agreements may include SLA language covering the June outage; legal teams should read Anthropic's suspension statement and the June 30 Redeploying Claude Fable 5 follow-up.
- Cache system prompts — prompt caching saves 90% on repeat input, which matters again now that Fable 5 is billing live. API guide
Cost comparison
| Model | Input / Output (per M tokens) | vs Fable 5 |
|---|---|---|
| Claude Fable 5 | $10 / $50 | — |
| Claude Opus 5 (July 24, 2026) | $5 / $25 | 50% cheaper |
| Claude Opus 4.8 | $5 / $25 | 50% cheaper |
| Claude Sonnet 4.6 | $3 / $15 | 70% cheaper |
Run your workload numbers on /pricing. For the full menu of replacements — including free and open-source options like GLM 5.2 — see our Claude Fable 5 alternatives guide.
Rebuild prompts on Opus
Fable 5's literal instruction-following still applies to Opus — but you may need to shorten XML blocks and lower effort for routine tasks. If you are picking a replacement rather than just handling a safeguard fallback, build against Opus 5 (claude-opus-5, same $5/$25 as 4.8); keep an Opus 4.8 variant for the classifier-rerouted sessions.
Start here: Prompt Generator · Template library · How to prompt Fable5 (patterns port across models)
What not to do
- Do not buy third-party "Fable 5 access keys." Official channels are listed on our access page.
- Do not assume jailbreak demos mean your API keys are compromised. The ban was a policy action, not a credential leak.
Bottom line
- Developers: re-enable
claude-fable-5, keep the circuit breaker and logging — the Opus 4.8 fallback is now a classifier behavior, not an outage workaround. - Subscribers: Fable 5 is included in up to 50% of weekly caps through July 7; after that, usage credits.
- Everyone: keep prompts sharp on the Prompt Generator.
Last updated July 2, 2026.