FinSkillBench is an evaluation suite designed to assess whether language model agents can effectively use financial domain skills for investment management tasks. The benchmark includes 12 subtasks across three domains: portfolio construction, risk management, and fundamental analysis. It compares the performance of models with and without access to curated skill packages, which consistently improve results. In contrast, self-generated skills provide little benefit despite increased computational cost. --- Why it matters: This matters because investment management is a high-stakes domain where AI agents must perform complex tasks beyond generating text. FinSkillBench provides a benchmark for evaluating the effectiveness of language models in using financial domain skills, which can inform model development and deployment in this critical area. Source: https://arxiv.org/abs/2608.18099