Over the last few days, a growing number of Claude users have posted the same claim: something new is answering when they pick Fable 5.1.

The outputs look sharper. Some answers seem to know about recent events without web search turned on. And the name people keep attaching to it is Fable 5.5.

Anthropic has not announced anything. There is no model ID, no docs page, no benchmark table. So this issue does what we do with every leak: separate what people are seeing from what anyone can actually prove.

Hidden routing. Users on claude.ai and in Claude Code say requests sent to Fable 5.1 are being answered by a different model. Their screenshots still show the Fable 5.1 label.

Fresher knowledge. One widely shared post from @kimmonismus said "queries without web search are now returning up-to-date results," and that "initial SVG tests are far superior to Fable 5.1."

Demos. X-ray animations, cartoons and voxel worlds are circulating as Fable 5.5 outputs. None come with server metadata proving which model made them.

A grey-test thread. User @chetaslua posted a seven-part thread describing early access. The claimed difference is that the model needs less direction and fills in improvements you did not ask for, like adding padding and highlights to a screenshot on its own. The thread also gave a rough order: Haiku first, then Fable 5.5 about a week later.

Here is the full list of things Anthropic has said in the last two weeks.

Opus 5.5 launched September 22, with Anthropic saying it performs at the level of Fable 5.1.

Sonnet 5.5 launched September 28.

Haiku 5.5 is "coming in the coming weeks." Anthropic said that in its own Sonnet 5.5 launch thread.

That's it. Fable 5.5 does not appear in Anthropic's model directory, API release notes or Claude Code model lists. Nobody has found a model ID, pricing or a single official benchmark.

We use the same three questions on every rumor. Here is how Fable 5.5 scores.

1. Has Anthropic done this before?

Partly. Routing rumors have shown up before every recent Anthropic launch. On September 18, users were sure they were being routed to "Fable 5.2." Something new did ship four days later, but it was Opus 5.5, not Fable 5.2. Before that, an "Opus 5.2" rumor traced back to a single account, and the model that actually shipped was Opus 5.5.

So the track record is consistent. Routing reports have been good at spotting that something is coming and bad at naming it.

2. Can you trace it to a document?

No. There is no code string, no config file, no leaked model card and no docs page. Everything so far is screenshots and outputs, and outputs cannot prove which model produced them. Even infrastructure changes can make the same model behave differently.

3. Has the date slipped yet?

Not yet, and this is the one to watch. The rumored order, Haiku first and Fable a week later, fits Anthropic's own "Haiku 5.5 in the coming weeks." If Haiku 5.5 ships soon and Fable follows about a week after, the rumor earns its credibility. If Haiku ships and nothing follows, it doesn't.

Scorecard: plausible, not proven.

Most of the evidence is vibes. One piece is more interesting.

If a model answers questions about recent events with web search switched off, it suggests the model was trained on newer data. That would point to a fresh pretraining run, not a tune-up of Fable 5.1. On September 10, one account predicted a new Fable pretraining run would land in "late September or early October." That is right now.

The honest caveat: caching, retrieval layers and other infrastructure changes can produce the same effect without new weights. It is a clue, not a confirmation.

Fable is Anthropic's top tier, and right now it is the oldest model in the lineup.

Opus 5.5 already matches Fable 5.1 at a lower price. Sonnet 5.5 beats Opus 5.5 on Terminal-Bench. And on Wednesday, Google's Gemini 4 Argon claimed the lead on most of its own benchmarks. A Fable 5.5 would be Anthropic's answer at the very top, and the timing would line up with the run-up to its IPO, which Bloomberg reports could start marketing the week of November 9. That last part is our speculation, not anything Anthropic has said.

For you, the practical rules are simple. Do not migrate anything on a rumor. If you run Fable 5.1 in production and your outputs suddenly change, log it, because that kind of record is exactly what turns a rumor into evidence. And give it two weeks. Haiku 5.5 is the tell.

The stories moving fast beyond today's headlines:

🏟️ LMArena is where unreleased models often show up first under codenames. If Fable 5.5 is real, this is one of the likeliest places to catch it early.

📏 Epoch AI Benchmarks tracks independent scores across frontier models, so you can check launch claims against something the labs did not write.

📝 LLM by Simon Willison is a command-line tool that runs the same prompt across models and logs every response to a local database. It is the easiest way to catch a model quietly changing under you.

Users say Claude is routing some Fable 5.1 requests to a new model they are calling Fable 5.5, with fresher knowledge and sharper output. Anthropic has confirmed Opus 5.5, Sonnet 5.5 and that Haiku 5.5 is coming. It has said nothing about Fable 5.5.

Past routing rumors have been right that something was coming and wrong about the name. Watch Haiku 5.5. If Fable follows about a week later, the leak was real.

Stay building. 🤖