Anthropic's Claude 5 Fable Shows Major AI Capability Leap
Original: What it feels like to work with Mythos
Why This Matters
Represents a significant advancement in AI capabilities and autonomous task execution
Ethan Mollick tested early access to Anthropic's Claude 5 Fable, part of the new Mythos-class AI models. The system outperformed all previous public models across tasks, working up to 12 hours on complex specifications and creating sophisticated games using only mathematical generation without external assets.
Mollick evaluated Claude 5 Fable across various tasks, finding it superior to existing public models by considerable margins. The AI created complex outputs including a sophisticated academic social science paper from a single prompt, a 10-page epic poem where every word starts with 's', and fully functional games using mathematical generation rather than external assets. In one test, Fable built an isochrone map showing travel times from various cities, incorporating airports, trains, walking, and driving data - a task no previous model completed successfully. The AI worked autonomously for multiple hours, researching thousands of potential trip distances and making numerous judgment calls. Mollick described the experience as both 'delightful and unnerving' due to the system's ability to execute complex requests independently.