Claude Fable 5.5, Explained Without Inventing It

Claude Fable 5.5 is not a model. It is a name with a claim behind it and nothing else: no identifier, no rate card, no page in any catalogue, no date. The claim was posted to X on the evening of October 2, 2026, which makes it five days old today, and that five-day-old post is the entire reason this page exists. Every model the claim measures against — Claude Opus 5.5, Claude Sonnet 5.5, Claude Fable 5.1, Claude Haiku 4.5, GPT-6 Astra and GPT-6.1 Sol — shipped more than seven days ago and would be unremarkable on its own. None of them is the subject here. The subject is the name at the front of the sentence, what is claimed for it, and what a reader who arrived from that claim can actually call today.

What the October 2 claim says, in full

The assertion is short enough to quote without compression. At 22:10 UTC on October 2, 2026, the account @kimmonismus quote-tweeted a post and wrote: "It's not really a surprise, but Fable 5.5 is expected to outperform Astra 6.1. Reminder: Astra 6.1 is already finished and was supposed to be released, but was postponed due to security concerns. Fable 5.5 is very close and rumoured to even outperform Astra 6.1." The post being quoted is @synthwavedd's, timestamped 22:00 UTC the same evening, and it reads in full: 5.5 > 6.1.

That is the chain — two posts, ten minutes apart — and it ends there. No console screenshot, no registry or catalogue entry, no staff comment, no repository commit, no job listing, no vendor communication. No parameter count, no context window, no benchmark result and no date. That is not a complaint about the sourcing; it is the shape of it, and it is why the useful work here is not on the rumour but on the models the rumour names. The sourcing itself, post by post and account by account, is audited in our October 3 write-up of the Claude Fable 5.5 leak, which is where the two-posts-deep reading belongs.

Five places a shipped model leaves a trace, and none of them has one

A model that has actually shipped leaves marks, and the marks are checkable this afternoon. All five of the following were read on October 7, 2026, and all five are empty of Claude Fable 5.5.

• The vendor's models overview — lists exactly four Claude models, with identifiers: Claude Fable 5.1, Claude Opus 5.5, Claude Sonnet 5.5 and Claude Haiku 4.5. There is no fifth entry, and no row is marked as pending.

• The vendor's platform sitemap — carries paths for fable-5 and fable-5-1 and stops. There is nothing after fable-5-1 to link to.

• Our own catalogue — the Anthropic line ends at claude-fable-5.1, claude-opus-5.5, claude-sonnet-5.5 and claude-haiku-4.5. Requesting a Fable 5.5 identifier returns no model. That is what a routing platform looks like when the model does not exist: there is nothing to route to.

• Artificial Analysis — no model page, and therefore no index figure for a Fable 5.5 in any configuration.

• Any rate card — no vendor price exists for an input token or an output token, so there is no number to budget against.

None of that is evidence against the rumour. It is evidence about the present tense, and the distinction is worth holding onto, because an unannounced model has no documentation by definition — an empty checklist is exactly what a true rumour would also produce. What the emptiness does tell you is how to read the page you just came from. A page showing a Fable 5.5 rate card, a context window or a benchmark score has produced something the vendor has not. If a produced-looking Fable 5.5 page or rate card turns up anywhere before the vendor publishes one, that page is the thing that is fake.

Why "beats Astra 6.1" is a strange sentence to write

On September 28, 2026, the Wall Street Journal reported that OpenAI had finished a model and decided not to ship it, because it regressed on safety rather than merely failing to improve. Wire coverage followed the same day, and the model was GPT-6 Astra. Before that decision, the UK AI Security Institute had run GPT-6 Astra through simulated cybersecurity scenarios with its cyber classifiers deliberately disabled, so the measurement captured what the model attempted rather than what its guardrails allowed. The Institute reported that GPT-6 Astra reached the most severe stage of the attack — delivering a malicious payload to an open-source maintainer's codebase — in 29% of all samples, against 6% for GPT-5.6 Sol and 0% for GPT-5.5, the last of those on a smaller seed set, and noted the figure is a lower bound because some runs exhausted the evaluation budget before reaching that stage. Every tool call was simulated; no real system was reachable. The report is at arxiv.org/abs/2609.38415. OpenAI released GPT-6.1 Sol the following day, September 29, positioned as Astra-like performance at a fraction of the price.

Read together, those facts make the sentence "Fable 5.5 outperforms Astra 6.1" name, as its benchmark, a model the vendor chose not to run. A cancelled release is not a slow release. Nobody can call GPT-6 Astra, so the comparison is not one a reader can reproduce, and the measurement that would settle it does not exist — which is what makes the claim durable in a way an ordinary benchmark claim is not. There is no result that could contradict it.

What to use now, and what the shipped models actually measure

Here are the five shipped models the claim implicitly ranks, each with the figure Artificial Analysis publishes for it on Intelligence Index v4.3 — those are Artificial Analysis's measurements, not ours — and the vendor list price. The v4.3 revision adds AutomationBench-AA, drops τ³-Banking and moves Terminal-Bench to 4.0; scores from earlier revisions of the index are not comparable with these.

• Claude Opus 5.5 (max with fallback) — 57.62 on v4.3, released September 22, 2026, $4 per million input tokens and $20 per million output tokens.

• Claude Sonnet 5.5 (max with fallback) — 56.00, released September 28, 2026, $2 and $10.

• Claude Fable 5.1 (max with fallback) — 53.35, released September 1, 2026, $10 and $50, with a 1M-token context window and up to 128K output tokens.

• GPT-6 Astra (max) — 52.67, released in September 2026, $10 and $50, with a long-context tier above 272K prompt tokens at $20 and $75.

• GPT-6.1 Sol (max) — 51.83, released September 29, 2026, $2 and $10.

Three gaps fall out of that list, and they are the part worth keeping. The compounding pair the rumour leans on — Claude Opus 5.5 over GPT-6 Astra — is 4.95 index points apart today, 57.62 against 52.67, and that is before any Fable 5.5 exists. The shipped Fable does not do that: Claude Fable 5.1 sits 0.68 points above GPT-6 Astra, 53.35 against 52.67, and 4.27 points below Claude Opus 5.5 while costing more than twice as much per output token. And Claude Sonnet 5.5, released six days after Claude Opus 5.5, sits 1.62 points behind it at 56.00 for half the output price. Our October 3 leak write-up works through that spread in full; the one-line version is that within a single Anthropic generation the index gap is wider than the entire distance between Anthropic's mid-tier and OpenAI's flagship.

That is what the October 2 claim is worth to a reader deciding what to call this month: not a forecast, but a reminder that the tier structure already produces the result the rumour treats as surprising. The claim is five days old today, and a concrete assertion that recent about which model to reach for is the only thing that puts an unreleased name on a useful page at all.

The Fable you can call today is 5.1

Anyone who searched for a Fable successor should be told plainly what the shipped Fable is. Claude Fable 5.1 is the current model in the line: released September 1, 2026, a 1M-token context window with up to 128K output tokens, 53.35 on Intelligence Index v4.3, priced at $10 per million input tokens and $50 per million output. Claude Opus 5.5 is $4 and $20 for 57.62 on the same revision — 4.27 points more capability for under half the input price and well under half the output price. That is not an argument against Claude Fable 5.1, which is built for a different posture, but it is the comparison to run before assuming the more expensive Fable is the one to reach for.

The cheap end of the same line repeats the pattern, so it is worth stating too. Claude Haiku 4.5 — $1 and $5 per million tokens, a 200K context window, released in October 2025 — is the newest Haiku that exists. Haiku 5.5 is a rumour in exactly the same register as Fable 5.5, with no identifier and no model page, and it is written about that way in our piece on the Claude Haiku 5.5 claim. The rule generalises: a name without an identifier is not something you can put in a configuration file.

What a reader can actually do with this

The models the claim measures against are all callable now, and none of them requires a decision about an unreleased name. Claude Opus 5.5, Claude Sonnet 5.5, Claude Fable 5.1, Claude Haiku 4.5, GPT-6 Astra and GPT-6.1 Sol are served through one OpenAI-compatible key on OrcaRouter, at the provider's list price with 0% markup passed through rather than added on top — so when a vendor changes a rate, the number you see moves the same day it moves on the vendor's own card. The catalogue runs to 200+ models, and the list is on the /models page.

• Try the pair the rumour is really about — Claude Opus 5.5 at $4/$20 and GPT-6 Astra at $10/$50 are the same endpoint and the same key, so an evaluation against your own traffic is a model-string change rather than a migration.

• Compose rather than choose — the routing DSL and model fusion let a panel of these models answer together behind one call, which is a way to get an aggregate read without holding a benchmark table in your head.

• Keep a new route on a leash — automatic failover across providers is what makes it survivable to point a production path at a model released last week, or at one released this afternoon.

What none of it depends on is Claude Fable 5.5. There is no identifier to put in a config, no price to budget against and no date to schedule around, and the honest position is that nothing a reader does this month should change because of a five-day-old post. The thing to watch is not a model name but the vendor documentation: for any model Anthropic releases, the models overview gains a row, a model page appears and a rate card follows, and the model is live in our catalogue at the vendor's own price the same day. Until that happens, the right verb for Claude Fable 5.5 is not "wait for". The models you can call are already better documented than the one you cannot.