We had to build new evals for Fable

Anthropic has released Claude Fable 5, a Mythos-class model, marking a significant step forward in analytical reasoning, especially in complex and messy data environments. This model outperforms previous iterations, such as Opus 4.7, by achieving high scores across various evaluative metrics like Analytical Hard, Semantically Modeled, and Semantically Unmodeled tasks. Fable 5 excels in intuitive analytics, communicating findings effectively, and applying a "golden workflow" that involves beginning analysis in the semantic layer before delving into raw data. Although it occasionally overthinks on simpler tasks, it offers significant improvements in handling complex, open-ended queries. The model is being rolled out in Hex, with admins required to enable model data retention due to Anthropic's safety monitoring protocols. While Fable can be resource-intensive, its enhanced capabilities make it suitable for high-stakes analytical work, prompting teams to weigh the cost-benefit for their specific needs.