AGIEval: leaderboard

20 tasks (18 multiple-choice, 2 cloze) drawn from human exams in English and Chinese: Gaokao, SAT, LSAT, lawyer qualification, civil service and math competitions, from Microsoft (2023); accuracy.

Source: arxiv.org.

Interactive version: theaggregate.ai/benchmark?slug=agieval · How It Works · Data refreshed daily, snapshot 2026-09-05.