Benchmark Almanac

University of Tokyo

We track 3 benchmarks that University of Tokyo made or helped make.

1 in use 2 with public code 2 with a leaderboard here

Benchmarks

Name Kind Released Best score Status
MMLU-ProX MMLU-Pro translated into up to 29 languages with expert review. Multiple-choice accuracy per language. Knowledge in many languages Mar 2025 87% In use
MMLU-ProX Lite Models answer 658 MMLU-Pro questions per language, the same items in each of 29 languages. Multiple choice accuracy. Knowledge in many languages May 2025 74.1% Not in cards
MM-OCEAN Models rate apparent personality traits in short videos, explain those ratings, and answer questions about observable cues. Evaluation reports rating accuracy, absolute error, cue-question accuracy, and model-judged explanation scores. Video understanding May 2026 Not in cards