Benchmark
MMLU-Pro
Test challenging academic knowledge and reasoning with harder multiple-choice questions.
Explore the original evidence
Explore the original evaluation, its methods, and reported findings.
Open evalCoverage
Reasoning-focused questions across academic subjects, with up to ten answer choices and trivial or noisy items removed from the original MMLU pool.
Methodology
Answer each question under a fixed prompting and reasoning setup and measure accuracy against the official answer key. Use the benchmark version and subject breakdown to interpret comparisons.