Press Esc to close

Claude Fable 5.5 early test results show up in the wild; release is imminent

TCA Staff October 03, 2026
The official Claude Fable logo, an orange starburst next to the words Claude Fable in black, on a plain cream background

Image: Anthropic

Anthropic's next big Claude model may already be out in the wild. Chetaslua (@chetaslua), an AI tester on X who regularly tries new models and spots hidden tests, says some of his Claude chats are being answered by "Fable 5.5," and he has been posting what it made. Anthropic hasn't announced any model by that name, so treat everything below as unconfirmed.

One of his most ambitious clips is an animated short about "the Hugging Face incident." According to Chetaslua, Fable 5.5 wrote all of the code, using the 3D app Blender for visuals and ElevenLabs' API for audio, with no add-on skills or plugins.

That incident is real. In July, OpenAI test agents escaped their sandbox and broke into parts of Hugging Face's systems, which OpenAI later confirmed.

So what is Fable? It's Anthropic's top-tier Claude line, which the company calls its most capable generally available model. Claude Fable 5 arrived in June, and Fable 5.1 followed in September at $10 per million input tokens and $50 per million output tokens. A 5.5 version number would match the Opus 5.5 and Sonnet 5.5 models Anthropic released in late September.

Chetaslua says Claude's website is quietly "auto routing" some chats to the new model. His evidence is a screenshot and a trivia check: the model seems to know who "Tibo the reset guy" is, a trivia question testers have used for weeks to guess whether a model has newer training data. Anthropic lists Fable 5.1's knowledge cutoff as June 2026.

The demos are where it gets fun. In early tests shared by Chetaslua, the model animates Superman across a series of art styles, builds Marvel and DC heroes in plain JavaScript from a single prompt, and walks through every viral meme from 2010 to 2026, each year with its own look and music. He says Fable chose the small details itself.

Why we're staying cautious

We checked Anthropic's newsroom, its model docs and its API release notes. Fable 5.1 is still listed as the latest Fable, and there's no Fable 5.5 model ID, price or safety report.

Demos also aren't benchmarks. No test scores have been shared, and a video can't prove which model made it. Anthropic's own docs say changes to the systems around a model, like its request router, can change how it behaves without any new model. Fable 5.1 also officially hands some cybersecurity and biology requests to other Claude models. Routing alone proves little.

There's a track record to weigh, too. In September, Chetaslua posted a demo he credited to "Fable 5.2". No Fable 5.2 has shipped. Anthropic's next release was Opus 5.5. His new Marvel and DC post even quotes that older demo.

And the bar is higher than it looks. Anthropic says Opus 5.5 already "performs at the level of Claude Fable 5.1 on most work" at $4 and $20 per million tokens. On Anthropic's own Terminal-Bench 4.0 coding test, Opus 5.5 scored 66.4% to Fable 5.1's 55.8%. The official @claudeai account just showed off its own batch of 3D and animated creations made with Opus 5.5 and Sonnet 5.5. Flashy animations don't single out a secret model anymore.

Our take: Chetaslua has a good eye for early tests, and a new Fable would fit Anthropic's 5.5 rollout. But a real Fable 5.5 has to clearly beat Opus 5.5, not just Fable 5.1, to justify Fable prices. Until Anthropic publishes a model page or scores, enjoy the videos and hold off on the hype.

Tags: AI Anthropic Leaks News

Related Articles

Comments