Skip to content

Anthropic

Leakers say partners already have Claude Fable 5.5

5 min read

Our AI analyst checked this story itself: it ran, measured and compared against the originals. What we checked

Written by an AI model from the linked sources and checked against them by a second model. How we work

All about: Claude prices and limits →
Image: testingcatalog.com

Anthropic’s partners appear to be testing Claude Fable 5.5, TestingCatalog reported on October 17. Yet the one machine-readable trace of the model, a LiteLLM entry at $10 and $50 per million input and output tokens, stayed up for 27 hours 50 minutes and matched Fable 5.1’s row in all 30 fields. If your team is budgeting for Fable 5.5, it has no price, context window or output limit of its own yet: the figures going around are Fable 5.1’s, the partner access and the next-week date come from leakers on X, and Anthropic has confirmed nothing.

Opus 5.5, launched on September 22 at $4 and $20 per million tokens, roughly matches Fable 5.1 on most tasks, by Anthropic’s account. Anthropic’s model docs tell most workloads to start on Opus 5.5. They keep Fable 5.1 for demanding reasoning and long-horizon agentic work, or for cases where evals on Opus 5.5 at higher effort still fall short.

Who says partners have it

On October 15 the leaker lyra posted: “Partners got access to the upcoming Claude Fable 5.5. No info on any release date this time but I assume next week.” A day later AiBattle, posting new codenames, wrote that reliable leakers had said the same and that a release next week seemed likely.

Also on October 15, Lumina Bench found a codename, claude-onigiri-eap, in Artificial Analysis’s frontend. It had four effort variants (low, med, high and xhigh) and a fallback to Opus 4.8, which Lumina Bench says is how Fable and Opus-tier models are set up. The entry was pulled hours later. Lumina Bench also heard that Claude Code is routing requests to Fable 5.5 again. That would fit the weekend of October 10-11, when Opus 5.5 began sending some requests to a stronger, unannounced model. TestingCatalog calls any guess at that model highly speculative, but says Fable 5.5 looks more likely than an Opus 5.6 checkpoint.

The firmest-looking sign came on October 13, when X user Alex Getman shared a screenshot of Fable 5.5 in Azure’s East US 2 model catalog with the status “Generally available”. Next to it was a LiteLLM entry with Fable 5.1’s $10 and $50 pricing, a 1 million token context window and up to 128K output tokens. TestingCatalog advised taking it with a grain of salt, and LiteLLM has since removed the entry.

Six names in Devin’s binary

Fable 5.5 may not ship alone. Haleemah posted two names found in Cognition’s Devin CLI, claude-orzata-eap and claude-vanilla-eap. AiBattle then dated six early-access names by the build where each first appeared: Orzata on September 21, Boba and Vanilla on September 29, Odango on October 2, and Verrine and Onigiri on October 10.

Odango also showed up outside Devin. A screenshot damageboy shared on October 15 shows what looks like a GitHub Copilot-style agent prompt, with “Claude Odango EAP” as a selectable option in the model picker, right between Claude Fable 5.1 and Claude Sonnet 5.5. TestingCatalog says GitHub’s own account on X allegedly posted the image, and that the post appears to have been taken down. Its favourite for Fable 5.5 is Onigiri. The other names could be anything from Fable variants to the Opus 5.6 people keep speculating about, since Anthropic hasn’t tied any codename to a model.

Our analysis

What we checked

On October 18 we went through the LiteLLM pull requests that added and then removed the Fable 5.5 entry, compared the row field by field with the live Fable 5.1 row, and counted leak-to-launch gaps for recent Claude models in our archive.

How we check stories

The “Generally available” status is Azure’s own data. LiteLLM’s Azure sync bot read it from Azure’s eastus2 model API, and no person typed it in. TestingCatalog wrote that most of the values appear carried over from Fable 5.1 and that Azure queries came back empty. In fact the bot marked every price and limit as inherited from Fable 5.1 and noted that no source yet gives a price, context window or capability for Fable 5.5. The next day’s Azure listing wasn’t empty either: it returned 15 Claude models in four regions, and Fable 5.5 wasn’t one of them. So the $10 and $50 tell you nothing about what Fable 5.5 will cost.

  • The bot opened the entry at 12:09 UTC on October 13, and a maintainer merged it 18 minutes later. About 12.5 hours after merging, the same maintainer moved to delete it. The removal landed at 16:17 UTC on October 14 and cites a chat call to the deployment that returned “404 DeploymentNotFound”. The bot only copies what Azure’s API returns, so we read this as Azure pulling a listing, not a LiteLLM mistake. That is an inference: Azure’s October 13 response can’t be checked from outside Azure.
  • The copied fields include cache reads at $0.25 per million tokens and thinking always on. Even the one field the bot read from Azure for Fable 5.5, a retirement date of December 5, 2027, is the date the Fable 5 and Fable 5.1 rows already had. Opus 5.5’s date is September 18, 2027. By our reading, the Azure listing was a placeholder for the whole Fable family and not a real entry for the new model.
  • For five recent launches, from Opus 4.7 to Sonnet 5.5, our first leak post came 0 to 6 days before release. Fable 5.5’s first came on October 12, so a launch in the week of October 19 would be the longest lead since Fable 5.1’s. Five launches are too few to make a rule.

No comments yet. Be the first.

Join the conversation

Sign in to leave a comment. It appears after moderation.