Anthropic releases Claude Fable 5, first public version of Mythos model for code generation

Anthropic has released Claude Fable 5, marking the first publicly available version of its Mythos model line. The release represents Anthropic's entry into extended-execution code generation, competing directly with OpenAI's o1 and other reasoning-focused models.

Extended execution capabilities

According to Ethan Mollick, an AI researcher at the University of Pennsylvania, Fable 5 can execute on multi-page specifications for up to 12 hours continuously. In testing, Mollick reports the model "outperformed basically every other public model I have used by a considerable margin."

The model operates through Claude Code, Anthropic's code generation interface. Pricing, context window size, and benchmark scores have not yet been disclosed.

Single-prompt game generation

Mollick demonstrated Fable 5's capabilities by generating multiple complete video games from single prompts:

• Snake: A Pac-Man-style arcade game with continuous movement and boundary death mechanics
• Strata: An exploration game set in subterranean tunnels with lantern-lighting objectives
• Duino: A walking simulator based on Rainer Maria Rilke's poetry cycle, displaying text passages as the player moves

All three games were generated as functional, playable applications from initial prompts without iterative refinement, according to Mollick.

Beyond gaming

Mollick also used Fable 5 to create an isochronic map—a visualization showing travel times between locations. He described the accuracy and detail as "arresting," though specific metrics were not provided.

Technical details unknown

Anthropic has not disclosed:

• Model parameter count
• Context window size
• Pricing per token
• Training data cutoff date
• Benchmark performance metrics
• Relationship between Fable 5 and previous Mythos development

The company has not confirmed whether Fable 5 uses chain-of-thought reasoning or other extended thinking mechanisms to achieve its multi-hour execution capability.

What this means

Fable 5's ability to generate complete applications from single prompts represents a significant compression in the software development cycle for simple projects. The 12-hour execution window suggests Anthropic is competing directly with OpenAI's o1 model, which also offers extended reasoning time for complex tasks.

However, without benchmark data or pricing information, it's impossible to assess Fable 5's performance relative to existing code generation models like Claude 3.5 Sonnet, GPT-4o, or Gemini 1.5 Pro. The quality of Mollick's generated games—described as having graphics that "aren't great"—suggests the model may prioritize functional code generation over polished output.

The release timing is notable: Anthropic is pushing into code generation as OpenAI, Google, and others race to dominate the AI-powered software development market.