The MULTIFINGROUND benchmark evaluates whether models can recognize accounting concepts and extract corresponding numerical evidence across English, French, and Greek. It utilizes FY2025 disclosures from 90 issuers to measure the gap between multilingual recognition and grounding.
HOW THIS AFFECTS YOU
●
builderIf building fintech tools, test against this gap to avoid numerical hallucinations in non-English contexts.
●
researcherUse this to evaluate the reliability of multilingual financial reasoning.