Claude Mythos 5.5 and GPT-6.2 Astra both surface, AI world may see a wave of flagship model updates in November

Just as November began, the AI world is buzzing again.

Over the past two days, Anthropic and OpenAI have successively leaked news of new models. Someone found the name Claude Mythos 5.5 in the source code of Anthropic's official website, and a developer even posted what appeared to be a 3D game and interactive works generated by Claude Fable 5.5.

On the other side, OpenAI's not-yet-announced GPT-6.2 Astra has also appeared in some users' Codex usage statistics and API response records, and some even claim to have successfully called it.

Clues about the two companies' new models appeared almost simultaneously, and combined with earlier rumors about Qwen4, GLM-5.5, and Hy4, the AI community has begun to speculate whether November will bring another round of concentrated flagship model updates.

Claude 5.5 Accidentally Exposed on Official Website—Is a New Model Coming?

The story begins with an accidental discovery on Anthropic's official website.

Late at night on November 1, tech blogger @kimmonismus posted on X that the name Claude Mythos 5.5, which had not yet been officially announced, appeared in the HTML source code of Anthropic's official website.

According to the screenshot, on Anthropic's Mythos product page, both the hidden H1 heading and the alt attributes of images contained the text "Mythos 5.5."

Subsequently, tech media outlet TestingCatalog also noticed the change and speculated that Anthropic might be preparing a new generation of models.

For readers unfamiliar with Anthropic's product line, it can be simply understood that Mythos mainly targets high-risk research areas such as cybersecurity and biology, with strictly limited access; Fable, built on the same underlying model, adds safety protections and serves a broader user base.

Therefore, the accidental appearance of the name Mythos 5.5 on the official website has led outsiders to speculate whether Fable 5.5, aimed at ordinary users, is about to debut.

It is worth noting that early this morning, developer @chetaslua further discovered that Mythos 5.5 had appeared six times in the official website's HTML, but the body introduction still remained at Mythos 5.1.

Later, another user, @pankajkumar_dev, said that Anthropic had already removed the web content involving Mythos 5.5.

Meanwhile, another clue about Fable 5.5 was also dug up.

As early as October 13, the model identifier azure_ai/claude-fable-5-5 had appeared in the Azure pricing configuration of the open-source large-model API gateway LiteLLM. The relevant configuration was subsequently removed, but the third-party MLflow model catalog still retained the corresponding record.

According to the circulating configuration information, Fable 5.5 may support a 1 million token input context and a maximum output of 128,000 tokens, at $10 per million input tokens, $50 for output, and $0.25 for cache reads.

As for when the new model will actually be released, the community is also divided, with some predicting this week, some next week, and others believing it will not be until mid-November, because it could be the best gift for Anthropic's IPO.

Claude Fable 5.5's Coding Ability Exposed—Can It Generate 3D Games from a Single Prompt?

Compared with a few characters in the official website's source code, what excites developers more are the Fable 5.5 demo videos that have been appearing one after another on X.

On November 1, developer @notjazii posted a comparison video of about 36 seconds, claiming that using the same prompt, Claude Opus 5.5 and Fable 5.5 were each asked to generate 3D scenes.

In the footage, Opus 5.5 mainly presents natural landscapes such as lakes, parks, fountains, and seasonal changes, overall closer to a miniature model display. The Fable 5.5 side, by contrast, shows seasonal changes on a farm, city buildings gradually growing, a harbor alternating between day and night, and even dynamic events such as volcanic eruptions and rockets launching.

Compared with static scenes, Fable 5.5's work emphasizes continuous animation, changes over time, and object states.

The poster believes the new model wins across the board in this comparison.

Soon after, he showcased a Minecraft-style game allegedly generated by Fable 5.5.

In a video of about 92 seconds, players can explore a blocky world from a first-person perspective, passing through forests, caves, coasts, and desert rivers, and can even see lava areas, pink-canopy forests, and dynamic lighting on the water's surface.

The game interface is also quite complete, including common elements such as a pickaxe, an item hotbar, and a categorized inventory. The author claims that the entire work was completed with just one prompt and took one to two hours to generate.

Two other demos are also quite noteworthy.

Developer @abhinavflac showcased an interactive historical sailing ship allegedly generated by Fable 5.5 through a zero-shot prompt.

The camera can move around the hull, observing the deck, gun ports, sails, and rigging structure, and the ocean waves, stern wake, and water reflections under the sunset are also rendered.

Another developer, @chetaslua, showcased a Three.js-based sports car application, where users can view the vehicle from different angles and switch paint colors at any time, including red, blue, silver, black, and fluorescent green.

Judging from these works, developers have begun trying to let AI take on more complex 3D modeling, scene animation, and interactive development tasks. However, the related videos do not include complete prompts, call logs, or source code, so the model identity, generation time, and whether it was completed in one pass still mainly rely on the authors' own accounts.

If this can be verified later, Fable 5.5's performance in complex interactive application development is indeed worth looking forward to.

GPT-6.2 Astra is here too—are Codex users getting early access to the new model?

Anthropic has been full of news, and OpenAI has also shown new activity.

On November 1, X user @TokenGremlin posted that in some users' Codex usage statistics, a model name that had not yet been officially announced appeared.

gpt-6.2-astra

According to screenshots from netizens, in one account's model usage breakdown, GPT-6 Astra accounted for about 80%, while GPT-6.2 Astra accounted for 19.8%.

Interestingly, users still see GPT-6 Astra in the model selection interface, but the backend statistics show the new model identifier.

Subsequently, developer @Mr_Salio also posted a piece of API response data, in which the response.model field was shown as gpt-6.2-astra.

The two clues come from usage statistics and response records respectively, leading outsiders to speculate that OpenAI may have already begun testing a new version of Astra.

Furthermore, in the early hours of today, user @ai_funss said that after upgrading Codex CLI, it was possible to specify the model name in the terminal and try to call GPT-6.2 Astra.

The screenshot they showed not only displayed the gpt-6.2-astra medium identifier, but also showed 1M window, that is, a 1 million token context window.

As for performance, two very different assessments have already appeared in the community.

@TokenGremlin believes that GPT-6.2 Astra has made obvious progress compared with the previous generation, especially in design, technical capability, and programming performance, and may even surpass Claude Opus 5.5 on some tasks.

But another user, @Da7_Tech, believes that the new Astra has not yet reached the level of Opus 5.5, let alone the rumored Fable 5.5.

It is worth mentioning that on October 31, OpenAI's official account also posted a video with the caption "There are many possible futures. Which one do you want to create?", which also sparked speculation about whether it was generated by a new-generation model.

Looking back at the past month, the update pace of global large-model vendors has clearly accelerated.

On October 28, Google DeepMind had just released Gemini 4 Carbon, and now Anthropic and OpenAI have successively had clues about next-generation models leak out. Domestic models such as Qwen, GLM, and Hy4 are also expected to release new versions.

Today's large-model competition has gradually entered a somewhat absurd stage.

Before a flagship model has even officially appeared, netizens have already found various traces through official website source code, backend records, and demo videos, and have even begun arguing in advance about who will be the next world number one.

As for whether these clues are traces unintentionally left by the vendors or preheating signals intentionally released, probably only the vendors themselves know.

When the real product finally arrives, how to surpass the imagination the internet has built for it in advance may be more difficult than defeating competitors.

This article comes from the WeChat public account "APPSO", author: APPSO, which discovers tomorrow's products, and is published by 36Kr with authorization.