“In a direct comparison by the speaker, Claude Fable-5 identified significantly more bugs in a codebase than either GPT-5.5 or Claude Opus using the same prompt.”