Claude Fable 5.1 and Opus 5.1: what is actually confirmed

Through the second half of August 2026, at least ten sites were running "Claude Fable 5.1 release date" pages, several of them full explainers for a model Anthropic had not announced. Most were put up by companies that resell model access, holding a keyword ahead of a launch nobody had scheduled. Follow the specifics they repeat and they narrow down to one Chinese-language report and a short run of posts on X.

This page does the opposite thing. It is a verification tracker with a date on it. Every line below says what the claim is, who made it, when they made it, and whether anyone has since confirmed it. Where a claim came true, we say so and note how close the original guess was. Where it did not, we leave it in the table instead of quietly deleting it, because a rumor tracker that erases its own misses is worth nothing.

Checked 3 September 2026 against Anthropic's model documentation, deprecation schedule, support pages and announcement post, all fetched that morning. The X posts quoted below come from an X API capture of 1 September, not pulled again since, which is why no engagement figures appear.

The tracker, as of 3 September 2026

What landed on 1 September

Anthropic's model page for Claude Fable 5.1 records a release date of 1 September 2026 and an API identifier of claude-fable-5-1. It lists a one million token context window, 128,000 maximum output tokens, adaptive thinking that is always on, and input and output pricing of $10 and $50 per million tokens, unchanged from Claude Fable 5. Prompt cache reads moved to $0.25 per million tokens, which the announcement puts at 75 percent below the previous rate. The deprecation schedule gives claude-fable-5-1 a status of Active, with a retirement commitment no sooner than 1 September 2027.

Alongside it came Claude Mythos 5.1, an equivalent model restricted to participants in Anthropic's Project Glasswing programme. Access runs through two verification programmes, one for defensive security work and one for life sciences, and the announcement says Mythos 5.1 is currently available only to a set of US organisations. That restriction is Anthropic's own framing.

For anyone paying Anthropic directly, the plan detail matters more than the specification sheet. Its support documentation, checked 3 September, puts Fable on every paid tier and none of the free one, then splits it by seat. Max plans and premium Team seats include Fable models and can spend up to half of their weekly usage limits on them at no extra charge. Pro plans and standard Team seats do not include Fable in plan usage limits at all and reach it through purchased usage credits, which is billing on top of the subscription. Standard Enterprise seats work the same way, where the organisation has switched credits on.

The announcement also carries benchmark figures. Anthropic reports Fable 5.1 at 52.6 percent on Terminal-Bench-Science against 24.7 percent for Fable 5, and 73.4 percent on CursorBench 3.2.0 against 70.5 percent. Those are vendor-published numbers on vendor-selected benchmarks, so read them as a company's own scorecard. No third party had published a Fable 5.1 evaluation when this page went up.

One detail matters for anyone tracking the naming scheme. Anthropic's models overview now files Claude Fable 5 under "Legacy models (still available)", while the separate deprecation table shows claude-fable-5 as Active with a retirement date no sooner than 9 June 2027. That same deprecation page defines the two words as different lifecycle states, Legacy meaning a model that will receive no further updates. Two Anthropic pages describe one model in different vocabulary on the same day, which is a fair warning against reading one vendor page and calling it settled.

What did not appear on 1 September is any model called Opus 5.1. The deprecation table records the lifecycle status of every publicly available Claude model, and claude-opus-5 sits there as Active with no 5.1 successor beside it. Claude Opus 5, released 24 July 2026, is still the current Opus.

One leak had first-hand evidence, and it never said "5.1"

The credible part of the story started on 23 August 2026 at 17:59 UTC, when the leaks account @Lentils80 posted that Anthropic was testing two models called claude-marshmallow-eap and claude-melon-eap, adding that marshmallow "performs better than melon and is better to talk to than Opus 5" while neither was "Fable level". Two other accounts, @ai_for_success and @0xLogicrw, repeated the codenames the same day without adding evidence of their own, so they add volume and nothing else. The -eap suffix, short for early access preview, matches the naming pattern that preceded the Opus 5 launch.

Read that post carefully. There is no version number in it. Nobody in the original leak said "Fable 5.1" or "Opus 5.1". Those labels were attached afterwards by other accounts, then hardened into headlines by sites that never went back to the source.

The attaching went badly. On 26 August 2026 at 12:37 UTC, @pankajkumar_dev wrote that "Marshmallow looks like a new Fable version" and "Melon looks more like an Opus checkpoint". On 28 August at 18:13 UTC, the same account wrote that "Melon is likely Fable 5.1, while Marshmallow is likely Opus 5.1", and added that Anthropic had reportedly picked Melon as the release candidate. Same account, same two codenames, opposite assignments, 48 hours apart, with no evidence attached to either post. The second guess sits closer to what shipped, in that a Fable release arrived first, and that is not vindication. A claim that inverts inside two days and then half matches an outcome has told you nothing you could have acted on.

One genuine Anthropic statement is caught up in all this, and it is worth reading properly. Thariq, an employee whose X bio reads "Claude Code @anthropicai", replied publicly on 22 August 2026 to complaints about Opus 5's behaviour: "Yeah I agree that Opus 5 is a really spiky model and we want our models to be consistent and warm and feel like Claude. We're working hard on this and it is a huge priority for us." That is an on-the-record acknowledgement from an actual employee that Opus 5 has a consistency problem the company is working on. Where you meet the line that an Anthropic employee confirmed a 5.1, this reply is what is being pointed at, and it names no version, no date, no price and no benchmark.

Following the loudest claims back to where they started

The two dates that drove most of the traffic to this topic did not come from a leak at all. "Fable 5.1 on 31 August, Opus 5.1 on 5 September" traces to the DataLearnerAI community release tracker, where they were the tracker's own projections. We could not reach that tracker page to re-check it, so treat the attribution as second-hand. What is checkable is that neither date ever had an Anthropic source behind it.

The clearest illustration is the post that spread the delay story. On 28 August 2026 at 20:47 UTC, @kimmonismus wrote: "The assumption was actually that Fable 5.1 and Opus 5.1 were supposed to be released yesterday. It seems to be delayed until next week." The word "assumption" is in the original. He was quote-posting an account suggesting readers interrogate their own Claude with trick questions to detect a new model, a method that produces confident answers and no evidence. That is the entire evidentiary basis for the delay, and the release it was supposedly delayed from had never been scheduled in public. Then 31 August passed with nothing, and Fable 5.1 arrived on 1 September. A tracker's guess missed by a day on one model, close enough to look prophetic in a screenshot and still no help at all on the second date.

The staged rollout claim travelled a different way. The story that a slice of Claude users had already been moved onto Fable 5.1 during August originates with the Chinese outlet 36kr, on 19 and 27 August 2026. The 19 August piece says Fable 5.1 had entered "grayscale testing", a literal rendering of the Chinese term for a staged rollout, and points at Fable 5's own June launch, where it says about 5 percent of users were picked for a quiet test first, as the precedent. English write-ups then credited TestingCatalog, an outlet with real reporting standing. TestingCatalog's Anthropic archive, checked on 3 September, carries exactly one Fable 5.1 story and it is the 1 September launch. A Chinese aggregator report picked up a Western citation it never earned.

The "racing OpenAI's Astra" framing deserves its own line, because it is the most repeated piece of unattributed context in the whole cluster. Astra is real. OpenAI disclosed the name of its next major model family, first reported on 1 August 2026, and on 7 August TechCrunch reported that the company had suspended work on some aspects of it, quoting OpenAI saying its "preliminary evaluations indicate strong enough performance that we cannot rule out Critical capability level at this time". That is a pause on internal work, and there is no launch date to move. Everything connecting Astra to Anthropic's release timing is somebody's inference, published without a source and without an Anthropic comment.

Where we stand, and how this page gets updated

Being straightforward about our own position: MultiChats does not carry Claude Fable 5, and we do not carry Claude Fable 5.1. Our registry, checked on 3 September 2026, serves seven Claude models: Opus 5, Sonnet 5, Opus 4.8, Opus 4.7, Sonnet 4.6, Sonnet 4.5 and Haiku 4.5, inside a catalogue of 60 models from 18 providers, 21 of them available on the free tier. If you came here looking for Fable access, we are not where you get it.

This page is built to be updated. When Anthropic publishes an Opus 5.1, that row moves to Confirmed with the announcement date attached, and the rows recording who guessed what stay where they are. A disproved claim stays too, marked. The value of a tracker is the record of what was said before anyone knew, which is why every line carries a name and a timestamp.

Two open items we will be watching. First, whether Anthropic ever acknowledges the -eap codenames, which would settle the marshmallow and melon question that a dozen posts guessed at. Second, whether independent evaluations of Fable 5.1 reproduce the gaps in Anthropic's announcement, since right now the only numbers in circulation are the vendor's own.

Our current Claude lineup and what each plan includes are on the MultiChats pricing section.

Last verified: 3 September 2026.