FINESSE-Bench: A Hierarchical Benchmark Suite for Financial Domain Knowledge and Technical Analysis in Large Language Models

Researchers introduced FINESSE-Bench, a hierarchical benchmark for evaluating large language models in finance across multiple skill levels.

The suite contains 3,993 questions from eight specialized datasets, including certification exams, trading tasks, and a Russian-language olympiad.

It uses a unified scoring protocol for multiple-choice, numeric, and free-response tasks, enabling more consistent comparison across models.

The benchmark helps reveal whether LLMs can do real financial work, not just basic finance question answering.