LiveK12Bench: New Benchmark Tests LMMs on Real High School Exams

Researchers have introduced LiveK12Bench, a dynamic benchmark designed to evaluate large multimodal models (LMMs) on authentic high school-level examinations. Unlike static datasets prone to contamination, LiveK12Bench comprises over 2,000 verified questions from the latest real-world exam papers in Mathematics, Physics, Chemistry, and Biology. The benchmark features an automated pipeline for continuous updates, aiming to reflect genuine testing environments. This addresses limitations of existing benchmarks that are often restricted in modalities, disciplines, and evaluation criteria. The work is published on arXiv under identifier 2605.26781.