A MULTIDISCIPLINARY ACADEMIC BENCHMARK
UlumBench.
University-level questions across academic disciplines.
Explore what models know—and where they fall short.
Explore the code Results snapshot · September 2026
6,167Evaluated questions
140Topic labels
10Question categories
500Constructed families
01 / MODEL RESULTS
The leaderboard
5 MODELS · 6,167 QUESTIONS
Full score table & model snapshots
| Rank | Model / snapshot | Overall | MCQ | Open response |
|---|
Merged baseline: Experiment 2 label-only MCQs + Experiment 1 non-MCQs. MCQs use exact scoring; other answers use GPT-4o-mini judging. One saved response per question and model.