# The next Claude model is very close

> We compared two archived copies of Anthropic’s site. The label changed from Mythos 5.5 to 5.1. What this tells us about Fable and the next Claude.


![Editorial Fable 5.5 poster over a collage of yellow texture, a sunset and a black-and-white eye.](/news/claude-release-cadence/fable55-cover.png)

Editorial cover by [HeyMark](https://heymark.ai/en/). 

**Updated · November 3, 2026.** The name **Mythos 5.5 appeared in the HTML of Anthropic’s own website**. We downloaded two archived copies of its Mythos page: the October 31 copy contains it; in the November 1 copy, the label says 5.1. That gives us something new to compare directly, alongside the release history we analyzed yesterday.

## What changed between the two copies of Anthropic’s page

### Before and after on Anthropic’s site

Two captures of the same Mythos page, saved by Internet Archive.

| Capture (UTC) | Exact label | Mentions of Mythos 5.5 |
| --- | --- | --- |
| [2026-10-31T21:49:26Z](https://web.archive.org/web/20261031214926/https://www.anthropic.com/claude/mythos) | `Claude Claude Mythos 5.5` | 6 |
| [2026-11-01T20:05:45Z](https://web.archive.org/web/20261101200545/https://www.anthropic.com/claude/mythos) | `Claude Mythos 5.1` | 0 |

**6 mentions = 3 fields + their duplicates.** One hidden heading and the alternative text of two logos. The data used to initialize the page repeats those three values.

- Hidden heading · h1
- Desktop logo · alt
- Mobile logo · alt

After replacing the label six times, the two downloaded HTML files are byte-for-byte identical.

[Inspect the comparison and hashes](/news/claude-release-cadence/mythos-html-comparison.json)

These are capture times; they do not establish when the site was changed.

The original phrase repeats “Claude”: `Claude Claude Mythos 5.5`. Both versions retain Mythos 5.1 content elsewhere on the page. The change could fit editorial preparation or a copy error; these files do not establish which explanation is correct.

To check it yourself, download the original HTML using the `rawUrl` links in our [verification file](/news/claude-release-cadence/mythos-html-comparison.json). Replace that entire phrase with `Claude Mythos 5.1` in the October 31 copy, then compare the files. That operation removes every difference between the two saved responses.

## Why Mythos also puts the focus on Fable

The connection comes from [Anthropic’s official FAQ](https://www.anthropic.com/claude/mythos): **Fable 5.1 and Mythos 5.1 share the underlying model**. Fable adds safeguards for general availability. This relationship exists independently of the release tracker.

It makes sense to watch Fable when a reference to another Mythos version appears. The possibility of both families receiving an update together remains our inference. The sources reviewed do not announce a Fable/Mythos 5.5 pair or provide a date.

## The LiteLLM reference that was already removed

The identifier `azure_ai/claude-fable-5-5` also circulated in LiteLLM’s cost registry. [PR 46428, merged on October 14](https://github.com/BerriAI/litellm/pull/46428), removed that entry from two files. The correction record explains that Azure did not list the model and that test calls returned **404 `DeploymentNotFound`**.

Those checks are documented by LiteLLM; we did not make authenticated Azure calls. The removed entry does not establish that Fable 5.5 was available. The archived HTML provides a separate trace from Anthropic’s own website, whose scope you can inspect above.

## What we can know about the next Claude today

On November 3, the [Mythos page](https://www.anthropic.com/claude/mythos) shows 5.1 again, and the [Claude Platform catalog](https://platform.claude.com/docs/en/models/overview) still lists `claude-fable-5-1`. These sources still provide no official date or published benchmarks for a successor.

The finding changes the available evidence: there is now a name observed in an archived copy of Anthropic’s site. The calendar below explains the earlier expectation, but does not turn that mention into a scheduled release.

---

## The tracker as we observed it on November 2

**On November 2, AI Release Tracker showed November 3 for Anthropic’s next model release.** When we opened its [company page in the browser](https://aireleasetracker.com/company/anthropic), the counter said one day remained. It also kept an earlier date: **October 28**, marked as five days overdue.

October 28 comes from recent release gaps. November 3 appears when the counter moves forward after that estimate has passed. We found no Anthropic announcement confirming November 3 in the official sources reviewed. The latest model we verified was **Haiku 5.5, announced on October 7**. [Haiku announcement](https://www.anthropic.com/claude-haiku-5-5) · [Official release notes](https://support.claude.com/en/articles/12138966-release-notes).

## Where October 28 and November 3 come from

The tracker takes the **five most recent gaps between release days**, calculates their median and adds it to the latest release date. Models announced on the same date count as a single event. It also moves the result to the nearest Monday–Thursday. Its [published methodology](https://aireleasetracker.com/expected) explains both rules.

For Anthropic, the calculation is reproducible: the gaps are **24, 39, 21, 6 and 9 days**. Sorted, they become 6, 9, **21**, 24 and 39. The middle value is 21; adding it to October 7 gives **October 28**, already a Wednesday, so no adjustment was needed.

A separate rule applies afterward. If that date passes without a recorded release, the site keeps the missed projection and displays the next occurrence of the weekday most common in its history. Here that is Tuesday, producing **November 3**. The tracker itself says this repositioning **does not improve the forecast**. It adds no leak, announced event or new information from Anthropic. [How missed dates are handled](https://aireleasetracker.com/expected).

### Claude: releases and tracker dates

| Date | Meaning | Source |
| --- | --- | --- |
| 2026-09-01 | Fable 5.1 | [Fable 5.1](https://aireleasetracker.com/model/anthropic/claude-fable-5.1) |
| 2026-09-01 | Mythos 5.1 | [Mythos 5.1](https://aireleasetracker.com/model/anthropic/claude-mythos-5.1) |
| 2026-09-22 | Opus 5.5 | [Opus 5.5](https://aireleasetracker.com/model/anthropic/claude-opus-5.5) |
| 2026-09-28 | Sonnet 5.5 | [Sonnet 5.5](https://aireleasetracker.com/model/anthropic/claude-sonnet-5.5) |
| 2026-10-07 | Haiku 5.5 | [Haiku 5.5](https://www.anthropic.com/claude-haiku-5-5) |
| 2026-10-28 | The tracker added 21 days to Haiku 5.5’s release: October 7 + 21 = October 28. That date passed without a newly recorded model. | [AI Release Tracker](https://aireleasetracker.com/expected) |
| 2026-11-03 | The tracker now shows November 3, one day after our observation. Once the first date passed, its rule moved the counter to the next Tuesday. Not an announcement. | [AI Release Tracker](https://aireleasetracker.com/company/anthropic) |

As of 2026-11-02; observed 2026-11-02T14:49:54.958Z. 26 days since the latest verified model; the ongoing wait is excluded from completed gaps.

### The five gaps used

| From | To | Days |
| --- | --- | ---: |
| 2026-06-30 · Claude Sonnet 5 | 2026-07-24 · Claude Opus 5 | 24 |
| 2026-07-24 · Claude Opus 5 | 2026-09-01 · Claude Fable 5.1 / Claude Mythos 5.1 | 39 |
| 2026-09-01 · Claude Fable 5.1 / Claude Mythos 5.1 | 2026-09-22 · Claude Opus 5.5 | 21 |
| 2026-09-22 · Claude Opus 5.5 | 2026-09-28 · Claude Sonnet 5.5 | 6 |
| 2026-09-28 · Claude Sonnet 5.5 | 2026-10-07 · Claude Haiku 5.5 | 9 |

Sorted: 6, 9, 21, 24, 39. Median: 21 days. October 7 + 21 days = October 28.

### Changing the sample changes the calculation

| Sample | Median (days) | Before weekday adjustment | Nearest Monday–Thursday |
| --- | ---: | --- | --- |
| Last 3 gaps | 9 | 2026-10-16 | 2026-10-15 |
| Last 5 gaps | 21 | 2026-10-28 | 2026-10-28 |
| Full history | 42 | 2026-11-18 | 2026-11-18 |

Illustrative HeyMark calculations using the tracker’s history, not new forecasts. Models on the same date are grouped. Weekday-adjustment ties go to Thursday. A different sample does not establish a more likely date.

### History used

| Release day | Models recorded |
| --- | --- |
| 2026-10-07 | Claude Haiku 5.5 |
| 2026-09-28 | Claude Sonnet 5.5 |
| 2026-09-22 | Claude Opus 5.5 |
| 2026-09-01 | Claude Fable 5.1 / Claude Mythos 5.1 |
| 2026-07-24 | Claude Opus 5 |
| 2026-06-30 | Claude Sonnet 5 |
| 2026-06-09 | Claude Fable 5 / Claude Mythos 5 |
| 2026-05-28 | Claude Opus 4.8 |
| 2026-04-16 | Claude Opus 4.7 |
| 2026-02-17 | Claude Sonnet 4.6 |
| 2026-02-05 | Claude Opus 4.6 |
| 2025-11-24 | Claude Opus 4.5 |
| 2025-10-15 | Claude Haiku 4.5 |
| 2025-09-29 | Claude Sonnet 4.5 |
| 2025-08-05 | Claude Opus 4.1 |
| 2025-05-22 | Claude Sonnet 4 / Claude Opus 4 |
| 2025-02-24 | Claude 3.7 Sonnet |
| 2024-10-22 | Claude 3.5 Haiku / Claude 3.5 Sonnet (upgraded) |
| 2024-06-20 | Claude 3.5 Sonnet |
| 2024-03-04 | Claude 3 Haiku / Claude 3 Sonnet / Claude 3 Opus |
| 2023-11-21 | Claude 2.1 |
| 2023-08-09 | Claude Instant 1.2 |
| 2023-07-11 | Claude 2 |
| 2023-03-14 | Claude 1 |

These are 30 tracker records, grouped into 24 dates and 23 gaps. They are not an official audit of every historical announcement.

[CSV](/news/claude-release-cadence/release-history.csv) · [Method and sources](/news/claude-release-cadence/method.json) · [AI Release Tracker](https://aireleasetracker.com/company/anthropic)

## The wait has exceeded the recent pace. How much does that tell us?

At the November 2 observation, **26 days** had passed since Haiku. That exceeds the recent 21-day median, but the median is not a deadline Anthropic committed to meeting.

A different sample changes the date. With only the last three gaps, the median falls to **9 days**: the raw sum lands on October 16, and the weekday adjustment moves it to October 15. Using all **23 gaps between distinct dates** in the collected history raises it to **42 days**, producing November 18. These are our calculations using tracker records, not alternative release dates. Open “See how it is calculated” to change the sample, or inspect the CSV.

This is why the counter needs context. The same history yields different dates depending on which portion we use. Passing one of them does not establish a verified probability that a model will arrive the following day.

## The calendar mixes families; it does not identify the next Claude

The six release days in the sample include **Sonnet 5, Opus 5, Fable/Mythos 5.1, Opus 5.5, Sonnet 5.5 and Haiku 5.5**. Official product notes confirm the recent sequence: Opus 5.5 on September 22, Sonnet on September 28 and Haiku on October 7. These are different tiers and versions in the lineup, not six complete generations. [Official product history](https://support.claude.com/en/articles/12138966-release-notes).

“Another Anthropic model” therefore does not identify **Claude 6**, another Opus or a Fable update. The [earlier speculation about Fable 5.5](/en/guides/claude-fable-5-5-release-rumors/) is a different kind of signal, with its own sources and limits; the calendar does not confirm that name.

Nor should every development carrying the Claude name reset this clock. **Duet**, announced on October 21, is a proactive agent app built on **Opus 5.5**. An update to the Claude Code client can also add features without changing the model. [What Duet is](https://claude.com/resources/articles/claude-duet) · [Claude Code changes](https://github.com/anthropics/claude-code/blob/main/CHANGELOG.md).

## What current Claude benchmarks show

If you are waiting before switching models, the date is only part of the decision. **There are still no published results for the next Claude in the sources reviewed.** We can look at what Opus, Sonnet and Haiku 5.5 already offer: coding, terminal use, reasoning and chart-reading tests measure different capabilities.

### The current reference: Claude 5.5

There are no published benchmarks for the next model in the sources reviewed. These are Anthropic-published figures for released models, not an independent HeyMark evaluation.

#### FrontierCode 1.1 · Main

Code changes evaluated for whether they could be merged without human edits.

| Model | % | Labeled setting |
| --- | ---: | --- |
| Opus 5.5 | 54.4% | Max |
| Sonnet 5.5 | 52.1% | Xhigh |
| Haiku 5.5 | 46.4% | Unverified |

Opus uses Max and Sonnet uses Xhigh; we did not verify Haiku’s effort for this score. Sonnet reports 46.2% at Max. More effort does not guarantee a higher score.

#### Terminal-Bench 4.0

Multi-step coding tasks inside a terminal.

| Model | % | Labeled setting |
| --- | ---: | --- |
| Opus 5.5 | 66.4% | Xhigh |
| Sonnet 5.5 | 70.6% | Unverified |
| Haiku 5.5 | 39.2% | Unverified |

Opus uses Xhigh and its announcement reports a ±2.6-point standard error. These are launch-table figures, not a HeyMark rerun with one agent and budget.

#### Humanity’s Last Exam

Multidisciplinary academic knowledge and reasoning, with tools.

| Model | % | Labeled setting |
| --- | ---: | --- |
| Opus 5.5 | 67.7% | With tools |
| Sonnet 5.5 | 64.5% | With tools |
| Haiku 5.5 | 57.4% | With tools |

All three results allow tools. They are kept separate from no-tools scores and are not a success rate for arbitrary tasks.

#### Chartography

Chart recognition and reasoning, without tools.

| Model | % | Labeled setting |
| --- | ---: | --- |
| Opus 5.5 | 64.4% | No tools |
| Sonnet 5.5 | 61.6% | No tools |
| Haiku 5.5 | 46.4% | No tools |

Opus’s 64.4% comes from the Sonnet table, without tools. The 89.0% in the Opus announcement allows tools: that is a different configuration and is not used here.

We do not calculate an overall score. Sharing a benchmark name and version does not guarantee an identical agent, budget, tool setup or safeguard policy. Small differences do not establish statistical superiority. Only verified settings are labeled; “Unverified” refers to effort. Opus uses production safeguards and may fall back to other models when they intervene.

[Opus · 22 Sep](https://www.anthropic.com/claude-opus-5-5) · [Sonnet · 28 Sep](https://www.anthropic.com/claude-sonnet-5-5) · [Haiku · 7 Oct](https://www.anthropic.com/claude-haiku-5-5)

Sonnet reports **70.6%** on Terminal-Bench 4.0 versus Opus’s **66.4%**, while the FrontierCode table lists **54.4%** for Opus at Max and **52.1%** for Sonnet at Xhigh. Settings are not matched. There is no universal winner: the task and reasoning effort change. These are results from Anthropic’s announcements, not our own tests or evidence about the model that comes next. [Sonnet table](https://www.anthropic.com/claude-sonnet-5-5) · [Opus table](https://www.anthropic.com/claude-opus-5-5).

## What would establish a release date?

Our November 2 review of the [Anthropic newsroom](https://www.anthropic.com/news), [product notes](https://support.claude.com/en/articles/12138966-release-notes) and [Claude Platform notes](https://platform.claude.com/docs/en/release-notes/overview) found no model announcement after Haiku 5.5.

An announcement identifying the model, availability and date would provide Anthropic’s information to plan around. The counter explains the expectation, but supplies no name, access terms or results for a future model.

*Update verified: November 3, 2026, 13:33 UTC. Comparison of two archived HTML responses and review of the linked sources. The tracker section preserves the November 2 observation at 14:49 UTC; its displayed date may change. HeyMark calculations and file comparison, with no internal Anthropic information.*

## Plan your next post in HeyMark.

Keep the idea, review the draft with your team, and see how it performed in the accounts you connected.

[Start free](https://app.heymark.ai) · [See how it works](https://heymark.ai/en/#product)

## Developer and agent resources

- [HeyMark MCP documentation](https://heymark.ai/en/mcp/)
- [llms.txt](https://heymark.ai/llms.txt)
- [Full site content for language models](https://heymark.ai/llms-full.txt)
- MCP protocol endpoint: `POST https://mcp.heymark.ai`
- OAuth protected-resource metadata: [/.well-known/oauth-protected-resource](https://mcp.heymark.ai/.well-known/oauth-protected-resource)
- MCP server card: [/server-card](https://mcp.heymark.ai/server-card)