Anthropic's Claude Fable 5 debuted last week as a safeguarded version of the Claude Mythos model, accessible to Claude Pro and Max subscribers. The author spent 72 hours testing the model before a U.S. government order suspended it. Despite the brief window, the hands‑on evaluation yields a clear picture of its strengths and shortcomings, and quickly attracted industry buzz.

In direct comparison, Fable outperformed Anthropic's Claude Opus 4.8 on several complex coding assignments. Tasks that required multi‑repo feature additions or intricate bug fixes, which Opus completed only after multiple prompts, were solved in a single shot by Fable. The model also generated thorough HTML reports uncovering severe security bugs and refactoring opportunities that Opus missed, demonstrating its potential for autonomous development.

The primary drawback lies in cost and token limits. Claude Fable charges $10 per million input tokens and $50 per million output tokens, quickly exhausting subscription quotas. Smaller firms find the pricing prohibitive, forcing a hybrid workflow where Fable handles planning and bug discovery while cheaper models like Opus perform routine implementation. The suspension leaves developers awaiting a stable release returns.