FINESSE-Bench measures LLM financial reasoning across eight certification levels

Large language models are moving into financial analysis, compliance, and professional training, but existing benchmarks don't capture the full range of professional competence—most stop at question answering over reports and skip the hierarchy of difficulty that defines real-world expertise.

FINESSE-Bench, released this week, is an eight-benchmark suite comprising 3,993 questions designed to test financial domain knowledge from foundational to expert level. The suite includes exam-oriented datasets inspired by professional certifications: CFA-like Levels 1-3, CMT-like Level 2, and CFTe-like Level 1, plus applied trading task collections and a Russian-language olympiad benchmark. The structure mirrors the progression practitioners face—basic concepts, intermediate quantitative reasoning, and advanced synthesis—and measures how models degrade as difficulty increases.

Widely used open benchmarks like FinQA, ConvFinQA, and TAT-QA focus on question answering over financial reports but don't provide an explicit difficulty hierarchy. Broader resources like FinanceBench, PIXIU, FinBen, and FLaME expand task coverage but leave the problem of evaluating the transition from foundational knowledge to expert-level reasoning open. FINESSE-Bench addresses this gap with a unified evaluation protocol covering multiple-choice questions, numerical answers, and short open-ended responses. The suite includes an automated scoring scheme for freeform answers based on the LLM-as-judge paradigm, enabling evaluation of domain breadth, performance degradation as difficulty increases, the ability to solve computational tasks, and model behavior in specialized financial domains.

The benchmark is intended both as a complement to existing open financial benchmarks and as a tool for more substantive evaluation of professionally relevant financial competencies in large language models.