Anthropic releases Claude Fable 5, its first Mythos model

Anthropic has released Claude Fable 5, marking the first publicly available version of its closely watched Mythos model family. The launch represents a significant milestone for the AI developer, introducing a model designed to handle complex, multi-step tasks directly from user prompts.

Ethan Mollick, an AI researcher and University of Pennsylvania scholar, tested the model and shared his findings on Tuesday on his Substack. Mollick reported that Fable consistently outperformed other public models he had used by a considerable margin. He found the model to be highly versatile, writing that it was “capable across many problems and produced some startling results — it would work up to a dozen hours executing on multi-page specifications.”

The model demonstrates significant capability in software creation, allowing users to generate functional projects from a single prompt using Claude Code. During his testing, Mollick used the tool to spin up several video games. These included a Pac-Man-like snake game where players navigate a screen to eat apples, and Strata, a game set in a network of subterranean tunnels where the objective is to light lanterns. He also generated Duino, an interactive experience based on the Duino Elegies by poet Rainer Maria Rilke, which features a lone figure walking through a nocturnal landscape as poetic passages materialize on the screen. While the graphics for some games, like Strata, resembled older, degraded styles, the fact that they were fully functional and generated from a single prompt highlights the model’s underlying code-generation capabilities.

Beyond gaming, Mollick used Fable to generate an isochronic map—a visualization showing travel time between locations—which displayed a high level of accuracy and detail. The ability to generate these diverse applications from a single prompt highlights the expanding toolkit available to “vibe coders,” a term used to describe developers who use AI to write code. Rather than relying on traditional development cycles, these users can now leverage models like Fable 5 to execute complex, multi-page specifications that previously required entire engineering teams.

Why it matters

Software projects that once required entire teams—including games, mapping tools, and highly complex specifications—are now being spun up from a single prompt. For founders and operators tracking AI capability curves, this release serves as a clear data point showing how quickly the floor is rising for automated software development.