MMLU-PRO from the Open LLM Leaderboard — enhanced version of MMLU with harder, multi-step questions across 57 academic subjects.
Runs the same MMLU-PRO task family under standardized settings for open models; wide subject coverage provides a broad academic reasoning signal.
Retired by Hugging Face on March 13, 2025; retained scores are historical. Leaderboards may differ in prompt and decoding setup, so cross-source numbers are not always interchangeable; multiple-choice format can mask reasoning failures through elimination heuristics.
Higher is better. Percentage accuracy.
Max score: 100
Retired by Hugging Face on March 13, 2025.
| # | Model | Score |
|---|---|---|
| 1 | 66.9 | |
| 2 | 63.9 | |
| 3 | 60.6 | |
| 4 | 60.0 | |
| 5 | 59.3 | |
| 6 | 58.7 | |
| 7 | 58.2 | |
| 8 | 58.0 | |
| 9 | 58.0 | |
| 10 | 57.5 | |
| 11 | 57.4 | |
| 12 | 56.8 | |
| 13 | 56.7 | |
| 14 | 56.7 | |
| 15 | 56.5 | |
| 16 | Gemini 2.5 Flash-Lite Preview (Sep '25) (Non-reasoning) source benchpress Evidence observed Apr 13, 2026 | 56.4 |
| 17 | 56.3 | |
| 18 | 56.2 | |
| 19 | 56.1 | |
| 20 | 56.0 | |
| 21 | 55.9 | |
| 22 | 55.7 | |
| 23 | 55.6 | |
| 24 | 55.2 | |
| 25 | 55.1 | |
| 26 | 55.1 | |
| 27 | 54.9 | |
| 28 | 54.8 | |
| 29 | 54.4 | |
| 30 | 54.3 | |
| 31 | 54.0 | |
| 32 | 54.0 | |
| 33 | 53.8 | |
| 34 | 53.5 | |
| 35 | 53.3 | |
| 36 | 53.3 | |
| 37 | 53.3 | |
| 38 | 53.1 | |
| 39 | 53.0 | |
| 40 | 52.9 | |
| 41 | 52.6 | |
| 42 | 52.4 | |
| 43 | 52.2 | |
| 44 | 52.2 | |
| 45 | 52.1 | |
| 46 | 51.8 | |
| 47 | 51.7 | |
| 48 | 51.7 | |
| 49 | 51.6 | |
| 50 | 51.4 |