Claude Fable 5.5 Leak Tracker: What Is Confirmed, What Is a Placeholder and What Not to Cite
A catalog entry, a paywalled headline and a few screenshots have set off Fable 5.5 buzz. Here is the evidence trail, sorted by what survives scrutiny.
Jump to 8 sections
Claude Fable 5.5 has no official release, price, model ID or benchmark. The evidence is a catalog entry that was added on October 13 and removed the next day, with a price copied from Fable 5.1, plus unverified rumors. Anthropic's latest confirmed Fable remains 5.1, so treat any 5.5 specification you see as a placeholder.
That has not stopped charts, screenshots and "benchmark" tables from circulating. Model Drop tracks new releases and what is actually different from the last one, and this roundup is for developers and analysts who need to know which Fable 5.5 claims are safe to repeat. We sorted every item we could find into confirmed, placeholder and unverified, and we show our work.
What Is Actually Confirmed About Claude Fable 5.5?
Almost nothing is confirmed. Kingy.ai's Fable 5.5 spec tracker, last checked on the evening of October 27, says Anthropic has not announced Fable 5.5 in the official sources it reviewed and that the latest confirmed Fable remains 5.1. The tracker lists the release date and time as unconfirmed.
What is real is the model family around it. Our coverage of the Claude Haiku 5.5 launch and pricing shows the 5.5 line is shipping, and Computerworld's price table lists Claude Sonnet 5.5 at $2 input and $10 output per million tokens. A Fable 5.5 would fit that pattern, but fitting a pattern is not evidence.
The last confirmed top-tier Fable is 5.1, released in early September 2026. Any claim that begins "Fable 5.5 scores" should be read as a claim about something that has not been publicly measured.
What Was the LiteLLM Catalog Entry?
The most concrete clue is a pull request in LiteLLM, an open-source library that keeps a price list for model providers. Per Kingy.ai, pull request 46201 was merged on October 13 and added an identifier, azure_ai/claude-fable-5-5. Its price, token limits and capabilities were inherited from Fable 5.1, and the tracker notes that no source established those values for Fable 5.5.
On October 14, a removal proposal said Azure did not list the model and no deployment existed. It was closed as superseded by a merged update, 46428. The tracker reads this as a corrected catalog, not as a release or a successful API call.
AIHubMix's roundup of the Kimi launch, which also tracks rumors, describes the same sequence and says the entry was removed a day after it appeared, with price, context and output limits copied from Fable 5.1. Those figures are $10 input and $50 output per million tokens, a 1 million token context window and 128,000 tokens of output.
Here is why that matters. Catalog entries are often created in advance, or by mistake, and a copied price is a placeholder, not a quote. Anyone who published "Fable 5.5 costs $10/$50" repeated a number that nobody at Anthropic has stated.
How Do the Rumors Stack Up?
The rumors have a poor record. Kingy.ai lists a September 10 post that predicted a late September or early October launch, an October 1 post that predicted an October 6 launch, and a September 18 post that used the label Fable 5.2. None of those dates produced a release, and the tracker says one poster later wrote that they could not verify any of it.
Several other claims cannot be tested. October 1 routing reports showed Fable 5.1 labels, which do not identify the model serving a request. An October 5 post claimed Claude Code was saving a rolling Fable selection, which the tracker could not reproduce. User demos credited to Fable 5.5 have unverified attribution.
Two items have more substance but still no proof. The Information reported on October 19 about a model with better intent understanding, according to the tracker, but the article is paywalled and gives no final name or date. AIHubMix adds that TestingCatalog reported on October 17 that partners have access, which is also unconfirmed.
The codename "Onigiri" has been floated, and a test model called claude-onigiri-eap was briefly seen on Artificial Analysis, per AIHubMix. Neither is confirmed, and a test label is not a product name.
The pattern here is familiar from earlier launches. Before a model ships, traces of it tend to appear in code repositories, partner programs and test leaderboards. Those traces are real, but they describe plans and tests, not products. Sorting them correctly is most of the job, and the cost of getting it wrong falls on readers who act on a number that never existed.
What Does the Evidence Table Look Like?
The table sorts each claim by what supports it. It is a snapshot as of October 27 and will change if Anthropic announces anything.
| Claim | Evidence | Status |
|---|---|---|
| Fable 5.5 exists as a product | No official announcement | Unconfirmed |
| Price is $10/$50 per million tokens | Copied from Fable 5.1 in a catalog entry | Placeholder |
| Context window is 1M, output 128K | Copied from Fable 5.1 | Placeholder |
| Partners have access | Report on October 17 | Unverified |
| Codename is Onigiri | Test label seen on a leaderboard | Unverified |
| Benchmark scores | None attributable | None exist |
Why Are There No Benchmarks Yet?
There are no benchmarks because there is no public model to test. The tracker says no evaluator has published attributable Fable 5.5 results, and the model has not appeared on Arena with a public score. Any chart that shows a Fable 5.5 bar was drawn from rumor or from a different model.
The only real data are for the predecessor. Kingy.ai notes that Fable 5.1 scored 55.8% on Terminal-Bench 4.0 in Anthropic's own figures and 58.08% in a run by Vals AI. That gap is a useful reminder that vendor and independent numbers can differ by a few points even for the same model.
Comparisons across time are also shaky. The tracker points out that Artificial Analysis changed its index methodology in version 5, so version 4 and version 5 scores are not directly comparable. A "new score vs old score" chart built across that change is misleading even when every number is real.
For a sense of what vendor-only claims look like once a model is real, see our GPT-6.1 Terra review, where every benchmark row is labeled by who reported it.
How Can You Verify a Release Yourself?
Check primary sources, in this order. First, look for an announcement on Anthropic's own site and a matching model page or system card. Second, look for the model ID in the provider's official model listing or API response. Third, look for the same ID in a cloud provider's catalog, with a working deployment you can call.
The checklist below applies to any model launch you read about.
- Is there an official announcement from the lab?
- Does the API or model list return the new model ID?
- Is there a system card or model card with evaluation details?
- Are the benchmark prompts and harness disclosed?
- Has an independent evaluator reproduced the headline numbers?
If an item fails the first two checks, call it a rumor. Our Reflection AI Beam launch review applies the same discipline to an open-weight release, and it is a good model for how to label what is known.
One more habit helps: record the date you checked. A tracker entry from October 27 says nothing about October 29, and a rumor can become a release overnight. Put the check date next to every claim in your notes, and re-run the same four or five checks whenever a new report appears.
What Should You Do While You Wait?
Keep building on confirmed models and avoid hard dependencies on rumored ones. If your product needs a stronger Anthropic model than Fable 5.1, plan a test using your own tasks the day a real model ID appears, and keep your prompts and scoring harness ready so you can compare quickly.
For writers and analysts, the rule is simple. Label anything about Fable 5.5 as unconfirmed, cite the source of every number and never publish a price or benchmark you cannot trace to the vendor. Attributing a copied price to Anthropic is the kind of error that is hard to retract.
It also helps to separate hype from planning. A rumor can be worth tracking as a signal of where labs are heading, such as better intent understanding in the next model, without being worth any spec in a table.
The Bottom Line on Fable 5.5
Fable 5.5 is a name with no public model behind it. The catalog entry was a placeholder that was corrected, the rumored launch dates have all passed and no benchmark exists. The reliable facts are about Fable 5.1 and the confirmed 5.5-family models.
At Model Drop, we will publish a full review the day a real Fable 5.5 model ID, price and system card exist. Until then, your next step is to bookmark Anthropic's announcements and model list, and to ignore any chart that shows a Fable 5.5 score.
