Where it stands
Anthropic has not announced Fable 5.5. Its release notes record Claude Haiku 5.5 as launched on October 7, and its models overview lists Claude Fable 5.1 but no Fable 5.5. A new outside X post from @AiBattle_ reports the names Verrine and Onigiri in the Devin CLI binary and says partners recently had access to Fable 5.5, with a release next week seeming likely; both points remain unconfirmed.
Outside account reports Verrine and Onigiri model names and partner access to Fable 5.5
An outside X account, @AiBattle_, says that new Claude early-access model names, "Verrine" and "Onigiri", have been spotted in the Devin CLI binary. The post says the names seem to have been added about a week before it was published on October 16, 2026.
The post also says that leakers it describes as reliable have reported that partners recently got access to Fable 5.5, and that a release next week seems likely. Anthropic has not announced either point in the material reviewed.
How this fits the official record
Anthropic's release notes, which run to October 15, 2026, record Claude Haiku 5.5 as launched on October 7. The notes do not mention Fable 5.5. The models overview lists Claude Fable 5.1, not Fable 5.5, as the model for demanding reasoning and long-horizon agentic work.
That makes the sequence in the earlier thread, with Haiku launching first and Fable afterward, harder to square with the official record. The Verrine and Onigiri names do not appear in the official pages reviewed.
Our read: a second outside account now points at the same unannounced model name, but it is still a leak with no official confirmation, so the story remains a rumour.
Anthropic Fable 5.5: What the Unverified Report Says
An unverified seven-post X thread describes an alleged Anthropic model called Fable 5.5 and attributes unusually capable task completion to it. In the thread, @chetaslua says they recently grey-tested the model, describing the experience as occurring less than 10 days before the post.
The thread says Fable 5.5 can complete a task from a short, one-line instruction without detailed prompting or reference material. It also suggests that Anthropic could launch the model after a planned Haiku release. The thread includes no official Anthropic confirmation or controlled evaluation of those claims.
What the thread reports about Fable 5.5
The central observation is that Fable 5.5 allegedly needs less direction than the author expected. According to the thread, a user can state the task in one line and the model will carry it out with what the author describes as expert-level attention to detail within the relevant subject.
That is a description of one person's experience, not a measured assessment of the model. The thread provides no test set, transcripts, benchmark results, task-completion rate or reproducible method. It also does not establish whether Fable 5.5 is an internal model, a test name, a product planned for release or a publicly available system.
The reported task-completion behavior
The thread focuses on a distinction between following an instruction literally and improving the result while completing it. As an example, the author says Fable took a screenshot of a Twitter page and added padding and highlights without being explicitly asked to make those changes.
If accurately described, that example would suggest the model inferred a useful presentation improvement from the task rather than merely reproducing the screenshot. But the thread doesn't include a transcript or other record of that example, so the result can't be checked. The report does not show how often this behavior occurs, whether the changes were appropriate or how the model performed on other tasks.
The author also argues that benchmarks may miss this kind of behavior because two models could receive similar scores even if one produces a more useful finished result. That is a general opinion about evaluation, not evidence that Fable 5.5 performs better than other models.
What the thread says about testing and comparisons
The account describes a recent, short grey-testing experience and presents it as an early impression. It compares the alleged model favorably with Anthropic's reported Opus 5.5 and makes broader comparisons with OpenAI models, but those comparisons are subjective. No controlled evaluation or supporting measurements are provided.
The thread also speculates that Anthropic's new pretraining could eventually lead to artificial general intelligence, or AGI—the hypothetical ability of an AI system to perform a broad range of intellectual tasks at a human-like level. That is the author's opinion, not a demonstrated capability or an established description of Fable 5.5.
What the thread suggests about a launch
One post says the plan was for Haiku to launch the following week and Fable to follow a week later. The wording presents this as a reported plan rather than a confirmed schedule, and the thread supplies no date, Anthropic announcement or product documentation.
The timing should therefore be treated as a prediction or leak. It does not show that either release occurred, that Fable 5.5 was ready for public access or that Anthropic had committed to that sequence.
What remains unconfirmed
The report does not establish:
that Anthropic has a model publicly or internally identified as Fable 5.5;
that the model has the reported task-completion behavior;
that the screenshot example occurred as described;
that Fable 5.5 outperforms Opus 5.5 or any OpenAI model;
that the model represents AGI-level progress; or
that Anthropic plans to launch it after Haiku on the suggested schedule.
More useful evidence would include an official Anthropic announcement or documentation, public access to the model, complete task transcripts, repeatable tests and independent evaluations. Until that evidence appears, the thread is best read as an early personal report about an alleged model—not as confirmation of a new Anthropic release or its capabilities.





0 comments
No approved comments yet. You can start the conversation.
Leave a comment