Finam AI Lab Updates Financial Benchmark FINESSE-Bench for LLMs

Finam Artificial Intelligence Lab has released an update to the FINESSE-Bench financial benchmark, designed to evaluate the knowledge and skills of large language models (LLMs) in the financial domain. Key changes include: fixing 169 problematic questions in the CFA-like Level 1 dataset; adding a new CFTe-like Level 1 dataset with 781 questions on the fundamentals of technical analysis; transitioning to confidence interval calculation via bootstrapping for more robust model comparison; using stratified bootstrapping to aggregate results by benchmark groups; separate assessment of dataset discriminative power and saturation (it was found that most FINESSE questions fall into the intermediate zone, providing good diagnostic value). The model pool has been expanded: a total of 66 systems were tested, of which 49 have full coverage across all 11 benchmarks (8 FINESSE + 3 public), and 33 models are included in the main tables. Top results were achieved by GPT-5.5 (0.906 on exam-like, 0.876 on trading/TA), Claude Opus 4.8 (0.860 on exam-like, 0.856 on trading/TA), and others.