We had to build new evals for Fable (hex.tech)

Anthropic has announced the release of Claude Fable 5, their first Mythos-class AI model that significantly enhances analytical reasoning capabilities compared to previous iterations like Opus 4.5. This new model excels in complex, messy data environments, scoring impressively across various analytical evaluations—93% on Analytical Hard and 93.5% on Semantically Modeled. These results highlight Fable 5's enhanced ability to navigate challenging analytical tasks that require a deep understanding of data context and nuances, marking a notable step change in AI-driven data analysis capabilities. What sets Fable 5 apart is its adeptness at effectively managing the "golden workflow," which combines semantic understanding with raw data analysis, ensuring accuracy while transforming complex queries into actionable insights. Moreover, its performance on the upcoming "Frontier" eval set, which tests open-ended and long-horizon analytical tasks, reveals its superior analytical depth and problem-solving skills. As Fable becomes available in Hex, admins must enable data retention for optimal use, although it may require more tokens for some tasks. This release not only offers a powerful tool for data teams but also sets a new benchmark for future AI models in analytical reasoning.