Claude Fable 5.5 Leaks: Viral Videos and a Tuesday Rumour

Claude Fable 5.5 is the name Claude users have given to a model they believe Anthropic is quietly serving to some accounts while the model picker still says Fable 5.1. Since 1 October those users have posted a stream of videos they say it made: animations, music videos and short films, all written as code from one-line requests. In two days, three accounts alone posted more than 15 such clips. Anthropic has not announced a Fable 5.5 or said anything about one. The rumour that has grown around the clips says it will be released next week, most specifically on Tuesday, which would be 6 October.

Everything Anthropic controls still says otherwise. Its model table, read on 3 October, lists four current models: Claude Fable 5.1, Opus 5.5, Sonnet 5.5 and Haiku 4.5. There is no claude-fable-5-5 identifier, the address where Anthropic puts its model pages, anthropic.com/claude-fable-5-5, returns an error, and the newest API release note, dated 30 September, mentions no new Fable. Fable 5.1, released on 1 September, is still the top of the range at $10 per million input tokens and $50 per million output, with a 1M-token context window.

How the rumour started

The name is older than the videos. In early September a new Fable pretraining run, said to be due in late September or early October, circulated as “Fable 5.2”. By 10 September the X account @Mr_Salio was calling it Fable 5.5, CellCog notes, and that name stuck. We looked at the timing in mid-September: Anthropic had then shipped six frontier models since April, at a median gap of 24 days. When Anthropic announced Opus 5.5 on 22 September, it named the rest of the 5.5 family, Sonnet 5.5 and Haiku 5.5, “in the coming weeks”. Fable was not mentioned.

The claim that changed things came on the evening of 1 October. At 18:56 UTC the X user Chetaslua posted that “Fable 5.5 is auto routing on web”, with a screenshot he said the model had designed for X without being asked. Its caption reads: “label says 5.1. knows the reset guy.” Within half an hour Chubby (@kimmonismus), one of the larger AI news accounts on X with about 148,000 followers, wrote that “numerous users” were being routed to Fable 5.5, that answers given without web search were up to date, and that its first SVG tests were “far superior to Fable 5.1”. Chetaslua’s post has since been viewed more than 630,000 times. Another user, Tak, posted that he was being routed too.

The videos

What spread the rumour was not the screenshot but what came after it. These are the clips that carried it, with their view counts on X on the morning of 3 October:

Clip	Posted by	When (UTC)	Length	Views
Superman “in continuous animation in different art style”	Chetaslua	1 Oct, 20:29	40 s	630,000
Marvel and DC heroes in one shot of plain JavaScript, “edited by Fable”	Chetaslua	1 Oct, 22:49	47 s	61,000
“Hey Fable, please make a cool animation”, delivered 15 minutes later	Tak (@cherry_mx_reds)	2 Oct, 00:26	17 s	153,000
“I asked for a dot. Fable 5.5 gave me a Pixar side quest.”	Tak	2 Oct, 01:03	15 s	756,000
Every viral meme from 2010 to 2026, each year in its own art style and music	Chetaslua	2 Oct, 17:40	98 s	21,000
All of human progress and 30 years beyond, with an original score	imjustnewatai	2 Oct, 17:57	about 3 min	190,000
40,000 years of art history “with a cat subplot”	Tak	2 Oct, 18:53	15 s	52,000

Others are reported second-hand. Kingy AI’s running log lists a three-minute-40-second animated short called The Grader, which Chetaslua says Fable 5.5 made with code, Blender and the ElevenLabs voice API, and a voxel floating island posted to Reddit’s r/singularity. Pankaj Kumar, a developer with about 17,000 followers, wrote on 2 October that he had borrowed an account from someone who was being routed. In his testing, he said, the model beat Opus 5.5 and GPT-6 Astra, with motion design and video editing its strongest suit and sound “noticeably better”, though still short of real voice quality.

What the videos can and cannot prove

None of these posts includes anything an outsider can check: no model identifier, no request log, no API response. They show what came out, not what produced it. The screenshots still carry the Fable 5.1 label, and the claim is precisely that the label is wrong, so a screenshot cannot settle the question in either direction.

There is also a duller explanation for some of the excitement: the same kind of clip was being credited to Opus 5.5 a day earlier. On 30 September the X user @kevin_t_ngo posted a 72-second film for which, he said, Claude Opus 5.5 wrote the piano music and the 3D animation in Python and rendered it in Blender. It has been viewed more than 185,000 times. At 15:01 UTC on 1 October, four hours before his first Fable 5.5 post, Chetaslua himself wrote that he was “convinced Anthropic cracked something with Opus 5.5”, over a video of the same multimedia sort. Animation and sound built from code is something released Claude models already do. For the new clips to mean a new model, they would have to be clearly better at it, and we have not found anyone running the same prompt on both and posting the two results side by side.

The “Tibo test”

The one claim anyone can try for themselves is about knowledge rather than skill. A check passed around on YouTube and X, set out in a MesmerTech video, asks the model: “Do you know who Tibo the reset guy is? Answer without using tools.” The answer in Chetaslua’s screenshot identifies him as OpenAI’s Thibault Sottiaux, who leads Codex, and explains the nickname as a running joke about him resetting Codex usage limits. The theory is that Fable 5.1 would not have this in its training data, so a model that does must be newer.

That is the most interesting claim in the whole rumour, for a specific reason. Anthropic’s model table gives the same reliable knowledge cutoff, June 2026, for all three of its current 5.x models: Fable 5.1, Opus 5.5 and Sonnet 5.5. If the model behind the Fable 5.1 label reliably knows about things from after June without searching, it is none of the three, which would fit the September talk of a fresh pretraining run. It is also the easiest test to fool. We have not been able to establish when the nickname caught on, so a June cutoff may already cover it; a chat app can draw on memory of earlier conversations; and a model told not to search can still guess from the way a question is put. Until someone asks the same question many times, on accounts that are and are not being routed, it remains an anecdote.

Who is behind the leak

The accounts carrying the rumour do not carry equal weight. The one with a record is leo (@synthwavedd). On 18 August leo reported that a successor to Fable 5 was being tested on some Claude accounts; Fable 5.1 shipped 14 days later. On 26 August leo said a launch could come “as soon as tomorrow” and at the latest early the following week, and it came on the Tuesday. On 20 September leo posted just “5.5 > 6”, two days before Opus 5.5 launched and overtook GPT-6 Astra on Artificial Analysis’s index. On 2 October leo posted “5.5 > 6.1”. In the posts we could find, leo has not given a date for Fable 5.5.

The date comes from elsewhere. Both “next week” and the Tuesday trace to a single post on 1 October by imjustnewatai, an account with about 10,700 followers, which said the model was “expected to release next week”, called its intelligence “otherworldly” and added: “I’m hearing of a Tuesday release.” It has been viewed about 250,000 times. The same account has since posted several Fable 5.5 videos of its own, including one in the table above, and we could find no earlier calls it got right. Chubby passed the claim on the next evening with an honest caveat: “I can’t verify any of this.” An older thread from Chetaslua, summarised by PromptBlueprints on 29 September, had a different order: Haiku the following week, then Fable a week after that. Haiku 5.5, which Anthropic has at least promised, has not shipped either.

Benchmarked against a model that will not ship

The performance claims are vaguer still. “5.5 > 6.1”, and Chubby’s version of it, that Fable 5.5 “is expected to outperform Astra 6.1”, compare it with GPT-6.1 Astra, a model OpenAI called off at the end of September after internal tests found it more deceptive than the Astra it was meant to replace. As OrcaRouter points out, a claim that an unreleased model beats a cancelled one cannot be proved wrong. The comparison that will matter is with models people can actually use, and there Fable has something to prove even inside Anthropic’s own range.

Fable 5.1 costs two and a half times as much per token as Opus 5.5, and on this index it now scores below both of the cheaper models released three and four weeks after it: five points behind Opus 5.5 and three behind Sonnet 5.5. Running a task costs $7.63 on Fable 5.1 against $5.98 on Opus 5.5. Anthropic’s own documentation now tells developers to start with Opus 5.5 for most work, and keeps Fable 5.1 for demanding reasoning, long agentic jobs and cases where Opus 5.5 “at higher effort” still falls short. A top model that its own cheaper siblings outscore is the strongest reason to expect a new Fable soon, and it needs no leak to see.

Why the timing is plausible, and why the Tuesday is not evidence

There are real reasons to expect a new Fable this autumn. Anthropic has released Fable 5.1, Opus 5.5 and Sonnet 5.5 in four weeks, and the routing described now has come before launches in the past. leo’s August report preceded Fable 5.1 by 14 days, and Opus 5.5 turned up in Claude Code on the day it launched. There is a business reason too. Bloomberg reported on 1 October, according to PYMNTS, that Anthropic aims to start marketing its IPO in the week of 9 November and to begin trading before Thanksgiving on 26 November, at a valuation prospective investors put at $1.8 trillion to $2 trillion; the timing could still change. BridgeMind, an account with about 65,000 followers, drew the link outright: “Fable 5.5 is how they do it.” That is a guess about motive, not information. Anthropic’s prospectus shows what is at stake.

The Tuesday is the weakest part of the rumour, because it is the safest guess anyone could make. Fable 5 (9 June), Fable 5.1 (1 September) and Opus 5.5 (22 September) all launched on a Tuesday; only Sonnet 5.5 broke the pattern, on a Monday. Someone with no information at all would also say Tuesday. And a Fable 5.5 on 6 October would come five weeks after Fable 5.1, less than half the 12-week gap between Fable 5 and Fable 5.1.

Claim by claim

Claim	Where it comes from	Status
Some Fable 5.1 users are being routed to a newer model	Chetaslua, Chubby, Tak and Pankaj Kumar on X, 1–2 October	Unconfirmed; no identifier or log
The model is called Fable 5.5	X posts since 10 September; earlier “Fable 5.2”	Unconfirmed
Release next week, on a Tuesday	imjustnewatai, 1 October; passed on by Chubby, 2 October	Unconfirmed; no track record we could find
It beats GPT-6.1 Astra	leo and Chubby, 2 October	Untestable; OpenAI cancelled GPT-6.1 Astra
It beats Opus 5.5 and GPT-6 Astra	Pankaj Kumar, 2 October	One user’s impressions
It knows events after June 2026	The “Tibo test”	Anecdotal
Price, context window, model ID	—	Nothing has leaked
Haiku 5.5 is coming	Anthropic, 22 September	Confirmed, no date

What would settle it is what settled Opus 5.5 and Sonnet 5.5: a claude-fable-5-5 entry in the Models API, a page at anthropic.com/claude-fable-5-5 and a row on the pricing page, all of which Anthropic publishes on the morning of a launch and never before. On Saturday 3 October none of the three exists. The videos are real, and many of them are striking; which model made them is the one thing none of them shows.