What it measures
- Knowledge breadth
- Reasoning under harder answer choices
- Academic subject coverage
MMLU-Pro
A harder MMLU-style benchmark for broad knowledge and reasoning.
How to read the score
% accuracy; higher is better.
Sourced Ranking
| Rank | Model | Provider | Score | Source | Source Type | Verified |
|---|---|---|---|---|---|---|
| 1 | Hunyuan-T1 | Tencent | 87.2 | Tencent Hunyuan-T1 documentation and release notes Accessed 2026-09-11 | official | Yes |
| 2 | DeepSeek-R1 | DeepSeek | 84 | DeepSeek-R1 model card Accessed 2026-06-11 | official | Yes |
| 3 | DeepSeek-V3 | DeepSeek | 75.9 | DeepSeek-V3 Technical Report Accessed 2026-09-11 | paper | Yes |
Limitations
- Confirm prompt format and model variant before comparing scores.
- The arXiv paper is static; the leaderboard and GitHub may update.
- HF Space is the score entry point; GitHub is the implementation entry point.
leaderboard
TIGER-AI-Lab GitHub and Hugging Face leaderboard
The arXiv paper is static; the leaderboard and GitHub may update. HF Space is the score entry point; GitHub is the implementation entry point.
paper
MMLU-Pro paper or specification
Use for benchmark definition, not necessarily for latest model scores.
third-party
MMLU-Pro dataset or code
Use for dataset, code, or implementation details; score freshness depends on the benchmark.