Leaked Anthropic Fable 5.5 Generates Animation via Pure Code in 2 Hours

On November 4 (Beijing Time), news emerged that Anthropic’s unreleased next-generation flagship model, Claude Fable 5.5, has recently garnered attention in overseas developer communities. Several developers shared early test results allegedly from the model on social media platform X. Among these, a 66-second animated short film generated entirely through programming stood out as particularly noteworthy. Testers claimed that without invoking any image or video generation tools, the model autonomously wrote and executed code for approximately 2 hours and 40 minutes to complete the animation visuals, background music, and sound effects.

On November 2, AI developer Chetaslua posted an animated short film on X, claiming it was independently completed by Fable 5.5. According to his description, the short is 66 seconds long with a resolution of 1920×804 and a frame rate of 24fps. The visuals include foxes, paper birds, clouds, text, and lighting effects, while the background music and sound effects were also generated by Python code written by the model.

Unlike AI tools such as Sora and Runway that directly generate video frames, this demonstration showcases a technical path for creating animations via code. Developers stated that no external images, asset libraries, audio samples, or font files were used throughout the process. The model autonomously handled scene design, visual rendering, and sound synthesis, ultimately packaging the final work into a single HTML file.

Beyond animation creation, several overseas developers have also demonstrated interactive 3D scenes allegedly coded by Fable 5.5, including ancient sailing ships, ocean waves, and sports cars. These demonstrations involve frontend graphics technologies like Three.js, where models, lighting, and object movements are controlled programmatically. Some scenes could be generated using simple prompts alone.

In other complex animation comparison tests, developers such as Jazil showcased results from different models generating bicycles, animal characters, and human combat scenes. Testers reported that Fable 5.5 could complete the coding and execution of complex dynamic scenes within about one hour, whereas some existing Claude models encountered issues like incomplete code and failed animation execution on similar tasks. However, these tests lack a standardized evaluation environment and cannot be directly interpreted as definitive model performance rankings.

Clues regarding the release of Fable 5.5 had previously surfaced multiple times. In mid-October, an entry named claude-fable-5-5 briefly appeared in the Azure directory of third-party model tool LiteLLM before being removed. Recently, developers discovered references to "Claude Mythos 5.5" in the HTML source code of certain pages on Anthropic’s official website, sparking speculation about an imminent new model release.

Under Anthropic’s current product structure, Fable and Mythos share the same underlying model but implement different levels of safety restrictions. Mythos is primarily targeted at vetted cybersecurity and life sciences research institutions, while Fable is made available to general users and developers with additional safety measures. However, Anthropic has not yet confirmed whether the 5.5 series will continue this design philosophy.

Coinciding with the exposure of the new model, a repository claiming to contain Mythos model weights recently appeared on the Hugging Face platform, with parameter counts reportedly reaching 14 trillion. However, researchers found clear signs of synthetic data in parts of the weight files, indicating there is no reliable evidence supporting claims of a model leak.

This wave of model leaks occurs during a critical phase as Anthropic advances its IPO plans. Prospectus documents disclosed by the company on November 2 show revenue of $33.242 billion for the first three quarters of this year. However, sustained increases in compute investment and significant losses have heightened external scrutiny on whether the next-generation model can deliver actual technological breakthroughs and commercial returns.

As AI model competition shifts from code completion to long-duration autonomous execution of complex tasks, the animation production and 3D scene generation capabilities demonstrated by Fable 5.5 reflect the potential application space for AI coding agents.

As of press time, Anthropic has not officially confirmed the release date, performance metrics, or availability scope of Fable 5.5. The aforementioned test results remain subject to official verification and independent evaluation.