Benchmark
FinanceBench
Grounded question answering about publicly traded companies using financial documents.
Explore the original evidence
Explore the original evaluation, its methods, and reported findings.
Open evalCoverage
The full benchmark has 10,231 questions. The original public evaluation used a 150-case sample and 16 model configurations.
Methodology
Evaluate answers and evidence for open-book financial questions; distinguish retrieval, long-context, and other configurations.