dailyai.report

XpertBench contains 1,346 tasks across 80 categories, drawn from more than 1,000 expert submissions. The domains span finance, healthcare, legal, education, STEM, and humanities, reflecting real‑world complexity. By offering rubric‑based evaluation, the benchmark gives developers a concrete measure of LLMs’ professional performance. Early tests show that current models still lag on specialized tasks, underscoring the need for targeted fine‑tuning.