共有:

Anthropic — Fable 5

The brain that went dark
is back in everyone's hands.

On June 12, an export-control reshuffle pulled Claude Fable 5 from every channel. Its stablemate Mythos 5 got a limited comeback for 100+ vetted US organizations on June 28, but general customers kept asking: when do we get Fable 5 again? In July, the door reopens.

AI Navigate Editorial2026.07.035 min read

Jun 12 All channels off Export controls Jun 28 Mythos 5, limited 100+ vetted US orgs Jul 03 Fable 5 back for all General availability
01

The Shutdown

June 12: Fable 5 vanished overnight

The export-control reshuffle took out the whole lineup, not just the top-end model.

On June 12, 2026, following a review triggered by revised US export controls, Anthropic pulled both Fable 5 and Mythos 5 from every channel. The Console API, Claude.ai, Bedrock, Vertex AI, and partner IDEs all stopped serving them. Teams that had built agent workflows around Fable 5 had to fall back to the Sonnet 5 line or the older Opus 4 line for the duration.

On June 28, the higher-tier Mythos 5 came back for 100+ vetted enterprise customers, but general-availability Fable 5 remained "no ETA." A quiet rumour even circulated that Fable 5 might just fold into the next-generation Fable 6 without ever returning as itself.

02

The Return

In July, the door reopened

After three weeks off, regular customers can call Fable 5 again.

21 days
off (Jun 12 → Jul 03)
All channels
Console / Claude.ai / Bedrock / Vertex
Clearer UI
Safety-routing now visible

On July 3, Anthropic announced that Claude Fable 5 is once again available to general customers. Three weeks after the shutdown and five days after the limited comeback, standard Console contracts, Claude.ai Pro and Team, and major cloud resellers all serve Fable 5 at the same pricing as before June 12.

Alongside the return, the "safety-routing" behaviour that quietly reroutes flagged prompts to a lighter model now surfaces itself in the chat UI. Instead of guessing why an answer came back short, you can see which model actually replied — a small UX change teams had been asking for.

03

Who It Matters To

Who is actually affected

Agent-ops teams

You can roll evaluation harnesses, prompt-tuning assets, and automations that were pinned to Fable 5 straight back onto it. For most teams this is a "revert to the pre-June-12 config" change, nothing new to build.

Individual / light users

You will barely feel it. If an answer suddenly comes back short, the new safety-routing indicator tells you which model actually replied, so you know whether to rephrase or escalate.

Procurement / IT

The risk of another regulatory shutdown hasn't gone away. Pairing Mythos 5's limited channel with Fable 5's general channel and keeping a use-case-based fallback plan is still the right posture.


Twenty-one dark days taught us to call
Fable 5 a non-substitutable asset.

AI Navigate — Daily Update · 2026.07.03