For six weeks, the AI internet has been chasing a model that does not officially exist.

Claude Fable 5.5 has turned up in an Azure catalog, in an open-source pricing database, behind secret codenames in Artificial Analysis and Cognition's Devin CLI, and in a GitHub model picker. One leaker says partners already have a second checkpoint. Prediction markets once put the odds of a launch by October 21 near 80%.

Anthropic has said nothing. Its models overview still lists Fable 5.1 as the newest Fable, and every predicted launch date has come and gone.

So today we are sorting the trail into what is confirmed, what is likely, and what is noise, and looking at the one date that might explain the timing: Anthropic's reported IPO in late November.

October 13: the Azure row. LiteLLM's price-sync bot added azure_ai/claude-fable-5-5, reportedly pulled from Azure's East US 2 catalog, and a screenshot showed it marked "Generally available." The listed specs ($10/$50 per million tokens, 1M context, 128K output) matched Fable 5.1 exactly, so they were almost certainly placeholders. LiteLLM removed the row when Azure came back empty.

October 15: the Onigiri codename. Lumina Bench spotted "claude-onigiri-eap" in Artificial Analysis's frontend, with four effort levels and an Opus fallback. It vanished within hours, then "claude-onigiri-eap-2" showed up two days later. Onigiri is the community favorite for Fable 5.5.

October 15: "may actually arrive today." Andrew Curran, who has a strong record on Anthropic timing, said Fable 5.5 might ship that day, while warning deployment issues could push it a week. It did not ship.

October 16–17: six codenames. Cognition's Devin CLI surfaced Orzata, Boba, Vanilla, Odango, Verrine, and Onigiri, early-access builds dated September 21 through October 10. A screenshot allegedly from GitHub showed "Claude Odango EAP" in a model picker between Fable 5.1 and Sonnet 5.5, and leaker Lyra said partners already had access.

October 19: a second checkpoint. A post claimed a second checkpoint had been delivered to partners. The same day, The Information reportedly said Anthropic was readying a model that reads user intent better, without naming it.

Short answer: almost nothing about Fable 5.5 itself.

Confirmed. Fable 5.1 shipped September 1 at $10/$50 per million tokens with a 1M-token context window. Opus 5.5 followed on September 22 at $4/$20 and, by Anthropic's own description, performs at Fable 5.1's level on most work. Sonnet 5.5 and Haiku 5.5 followed, and Haiku 5.5 on October 7 is still Anthropic's latest model announcement.

Unconfirmed. That Fable 5.5 exists under that name, plus its model ID, price, benchmarks, and launch date. Test requests to a claude-fable-5-5 model return a 404, and the rumored docs page shows "Page not found."

Mostly noise. Claims that Fable 5.1 or Opus 5.5 users are being silently routed to a stronger model. One tester ran the same "knowledge probe" against the shipped Fable 5.1 and got a false yes. A viral "Fable 5.5 is here" post ranks well in search with nothing behind it.

The pattern that matters. Companies do not hand six early-access codenames to partners, appear in a cloud catalog, and show up in client code for a model that is not coming. The question is when, not whether.

The best explanation for the delay may be on Wall Street, not in the lab.

Bloomberg has reported Anthropic is targeting an IPO as soon as mid-November, with marketing possibly starting the week of November 9 and a debut around November 26, at a valuation discussed between $1.8 and $2 trillion. Reuters separately reported, citing three sources, that Anthropic was considering shipping a new model before going public. Those dates could still move.

A flagship launch just before an IPO roadshow is the strongest demo a company can bring. It would also explain the long partner testing: a model that has to be bulletproof during an IPO window gets more checkpoints, not fewer.

Prediction markets have chased the leaks. According to Kie.ai, Polymarket priced a new Fable by October 21 at 79.5%. That date passed, and attention has moved to November.

What would Fable 5.5 need to do? Opus 5.5 already matches Fable 5.1 on most tasks at less than half the price, so a new Fable has to open a clear gap above Opus to justify a premium tier. Google's Gemini 4 Carbon, which took first place on the Artificial Analysis Intelligence Index this week, just raised that bar again.

Fable 5.5 is the worst-kept secret in AI, but it is still a secret.

The evidence for an imminent model is strong: codenames in third-party tools, partner checkpoints, a cloud catalog row. The evidence for any specific spec is weak, since every leaked number was copied from Fable 5.1.

Four things to do while you wait:

Build on Opus 5.5 today. It matches Fable 5.1 on most work at $4/$20. Don't park a project for an unannounced model.

Ignore leaked specs. The Azure pricing and context window were placeholders. Wait for the real rate card.

Abstract your model ID. When a new Fable lands, switching should be a config change, not a refactor.

Watch Anthropic's own pages. The newsroom and the models overview are the only sources that settle this.

The stories moving fast beyond today's headlines:

Spotted everywhere. Claude Fable 5.5 has surfaced in an Azure catalog row, in Artificial Analysis as "Onigiri," in six early-access codenames in Devin CLI, and in a reported GitHub model picker.

Confirmed nowhere. Anthropic's models overview still lists Fable 5.1 as the newest Fable, API requests for Fable 5.5 return a 404, and every predicted October date passed without a launch.

Leaked specs are placeholders. The $10/$50 pricing, 1M context, and 128K output were copied straight from Fable 5.1.

November is the date to watch. Bloomberg reports Anthropic is targeting an IPO as soon as mid-November, and Reuters reports it considered shipping a new model first.

What to actually do. Build on Opus 5.5 now, keep your model ID swappable, and trust only Anthropic's own pages for the launch.

Stay building. 🤖