Skip to content
AI Primer
update

Anthropic reports Claude Fable 5 sightings were a UI bug; traffic stayed at zero

After Bedrock cards, Claude Code strings, and app pickers suggested a return, Anthropic said Fable 5 was serving zero traffic and the sightings were a UI bug. That leaves visible IDs and client strings, but no production model access to route against.

4 min read
Anthropic reports Claude Fable 5 sightings were a UI bug; traffic stayed at zero
Anthropic reports Claude Fable 5 sightings were a UI bug; traffic stayed at zero

TL;DR

You can read the original Anthropic suspension notice, inspect the AWS Bedrock resurfacing post, compare the Claude Code string changes, and watch how model-self-identification screenshots briefly outran the actual backend state.

Zero traffic

The core fact arrived in one line. According to the zero-traffic reply, Anthropic was serving "exactly 0 traffic" to Fable 5 and suspected a UI bug.

That statement quickly collapsed the comeback narrative. Kimmonismus's correction, TestingCatalog's summary, WesRoth's follow-up, and AI Builder Club's recap all converged on the same point: visible model surfaces did not mean a production route was live.

Bedrock listings

The strongest false-positive signal came from AWS. The Bedrock listing post said Fable 5 had reappeared in AWS docs, model cards, and data-retention docs, with IDs including anthropic.claude-fable-5 and global.anthropic.claude-fable-5.

A separate Bedrock screenshot in the access-denied test mattered more than the card itself. It showed anthropic.claude-fable-5 in the interface, but the actual invocation path returned AccessDeniedException, which fits the broader pattern: catalog visibility, no confirmed serving.

Claude Code strings

The second signal came from client strings, not traffic. The Claude Code string report showed v2.1.190 adding or rewording Fable 5 copy, and the follow-up summary extracted the concrete changes.

The new strings pointed to a more specific billing and quota model than earlier builds:

Those strings are still evidence of preparation, not rollout. They tell you the client had logic for limits and weekly included usage, but the zero-traffic reply says there was no live traffic behind it.

What users actually saw

The user reports were not invented. The model-picker screenshot showed Fable 5 inside Claude's selector with a "Currently unavailable" label, and the iOS recording post claimed the model worked only in selected chats.

The more confusing evidence was the self-identification screenshot, where a chat answered, "I'm running as Claude Fable 5." After Anthropic's reply, posters like the red-herring follow-up and the later recap treated those outputs as symptoms of the same UI-state bug, not proof that requests were actually routed to Fable 5.

The corrections spread slower than the screenshots

This story moved in the usual bad order: first the sightings, then the caveats, then the debunk. An early rollout claim cited video proof, a later correction added Anthropic's message back into the thread, and a cleanup reply said the earlier AWS post had been deleted after direct confirmation.

That left one durable engineering takeaway: several surfaces could mention Fable 5 at once, including AWS cards, Claude Code strings, and app pickers, while the backend still served nothing. The official suspension notice at Anthropic's access page stayed relevant the whole time because the visible IDs and UI hooks never turned into confirmed invocations.

Further reading

Discussion across the web

Where this story is being discussed, in original context.

On X· 6 threads
TL;DR5 posts
Zero traffic4 posts
Bedrock listings2 posts
Claude Code strings1 post
What users actually saw3 posts
The corrections spread slower than the screenshots3 posts
Share on X