MMLU
UC Berkeley · v2024
Details
- Purpose
- Multi-task language understanding across 57 subjects
- Domain
- General knowledge
- Health
- Saturating
- Reproducibility
- Reproducible
- Contamination concern
- Medium
- Saturation
- Near-ceiling for frontier models
- Code available
- Yes
- Last reviewed
- 2026-06-20
Benchmark run records and model comparisons will be added in Phase 4.