Anthropic is preparing a model that understands what you want even when you barely explain it. The Information reported this on October 19, citing a person who has already worked with the model. There is no name, release date or benchmark yet, and Anthropic itself has announced nothing. Leaks from the past week call the model Claude Fable 5.5, and Polymarket traders are betting it ships before the end of October.
Let's sort out what is fact and what is rumor.
What the model can do, according to the tester
According to the tester, the model has noticeably stronger intuition than Anthropic's current commercial models: even from a sparse prompt, it works out what problem the person is really trying to solve.
The example from the article: you ask it to fix one bug in your code, and the model finds a whole class of problems that caused that bug and offers to fix the entire chain. This is exactly what today's coding agents lack: they dutifully patch the line you pointed at and miss three identical ones in neighboring files.
The second example is about unconventional solutions. Ask for advice on investing in stocks, and the model won't repeat the usual tips about diversification. Instead it will offer to write its own forecasting algorithm tailored to your goals. That could be a real find or confident fantasy: only real examples will show how good such an algorithm is.
The main caveat: all of this comes from one person. There are no numbers and no independent tests yet.
Why people call it Fable 5.5
The Information names neither the model nor the product line: it could turn out to be a new Opus or the next Fable. The name Fable 5.5 comes from leaks over the past week:
- October 13. A post on X: the author was "told" Fable 5.5 would ship that week. It didn't.
- October 14. ShiftDelete found the name Fable 5.5 in an Azure entry in LiteLLM, an open-source library with a directory of model prices and limits.
- October 15. Insiders report that Anthropic's partners already have access. For a few hours,
claude-onigiri-eapshows up in the Artificial Analysis interface with four effort levels (low, medium, high, xhigh) and a fallback to Opus 4.8. Fable 5 also hands some requests over to Opus 4.8, which is why onigiri is considered the main candidate. - Before October 17. A screenshot of the Azure catalog appears with Fable 5.5 marked "Generally available." The price and limits in the LiteLLM entry match Fable 5.1 exactly: $10 and $50 per million tokens, a 1M-token context and responses up to 128K tokens. The entry was removed a day later.
- October 17. TestingCatalog pulls it all together and writes that Anthropic will release Fable 5.5 next week.
None of these signals comes from Anthropic. The Azure figures were most likely copied from Fable 5.1, so they say nothing about the new model's price or context. The name itself makes sense, though: Opus, Sonnet and Haiku all moved to 5.5 in September, while Fable is still on 5.1. An Opus release isn't ruled out either: leaks link some of the test-model codenames found over the past month to a hypothetical Opus 5.6.
What we know for sure: The Information published its report, and the newest Fable in Anthropic's documentation is still 5.1, released on September 1, 2026. The new model's name, release date, price and benchmarks are unknown.
What Polymarket thinks
Polymarket has a market on the release date of the next Fable: any version after 5.1 counts, whatever it ends up being called. About $180,000 has been traded in it. The odds as of noon on October 20:
| Release date | Probability |
|---|---|
| October 20 | 12% |
| October 22 | 11-13% |
| October 27 | 15% |
| other days from October 21 to 30 | 2-4% each |
| not released by October 31 | 26% |
So traders give October roughly three chances in four. Over the last few days, the bet on October 20 climbed to about 40% and dropped to 12% by noon. I wouldn't try to guess the date from these lines: traders are working off the same leaks listed above, and some days have almost no money on them. October 30, for example, has only $427 riding on it.
With Fable 5.1, the leaks also got ahead of the release. The test model Honeycomb EAP flashed up in Cursor on July 8, and Anthropic released Fable 5.1 on September 1, almost two months later. More on that story in our roundup of the Fable 5.1 rumors.
Why now: Anthropic is preparing for an IPO
The timing of the leaks lines up with Anthropic's preparations to go public. On June 1, 2026, the company confidentially submitted a draft prospectus (Form S-1) to the SEC, right after a Series H round at a $965 billion valuation. According to Reuters on September 4, the listing will take place a few days before the US midterm elections in early November. Investors are reportedly discussing a valuation of around $2 trillion.
On September 28, Reuters obtained the draft prospectus. Anthropic's 2025 revenue was $4.6 billion with an operating loss of more than $8 billion, but in the second quarter of 2026 revenue grew to $11.5 billion, up from $4.73 billion in the first. The same day, Reuters put the link between releases and the listing right in its headline: Anthropic rolls out a second Claude 5.5 model "as it builds toward IPO."
On a roadshow, a company sells investors a growth story, and a new flagship a couple of weeks before the offering is more convincing than any promise in a prospectus. If the Reuters timeline holds, the end of October that Polymarket is betting on falls in the final weeks before the listing. Anthropic doesn't comment on such a link; this is our reading of how the dates line up.
What Google and OpenAI are preparing
Anthropic is not alone in this race. In the same report, The Information writes that Google is preparing a model codenamed Gemini 4 Carbon, and OpenAI plans to release a new mid-priced model in the coming weeks, possibly from the Terra line. In the outlet's view, fall 2026 will be a period of fierce competition among the three companies.
Where to try it
The new model isn't on GPTunneL yet. As soon as Anthropic releases it, it will appear in the same chat as the other Claude models, and we'll update this article.
For now, there's Claude Fable 5.1 for hard tasks: $0.02 per 1K input tokens and $0.1 per 1K output tokens. For everyday work, Claude Opus 5.5 at $0.012 and $0.06 is enough. One balance, no subscription, and switching models is a pick from a list.
Sources
- Anthropic Readies Model That Reads Users' Intentions Better, The Information, AI Agenda, 10/19/2026 - report on the new model.
- Anthropic prepares to release Claude Fable 5.5 next week, TestingCatalog, 10/17/2026 - roundup of leaks: partner access, Azure,
claude-onigiri-eap. - Claude Fable 5.5 is expected this week, ShiftDelete, 10/14/2026 - the October 13 post and the LiteLLM entry.
- Anthropic IPO launch shifts toward mid-October, Reuters, 09/04/2026 - listing timeline.
- Anthropic's $2 trillion IPO prospectus has leaked, Fortune, 09/29/2026 - revenue and losses from the draft prospectus.
- Anthropic rolls out second Claude 5.5 model as it builds toward IPO, Reuters, 09/28/2026 - a model release in the context of the IPO.
- Polymarket, market on the release date of the next Fable model (5.2+), 10/20/2026 - release probabilities by day.
- Claude models, Anthropic documentation - current models and their identifiers.