MATH Level 5 (hardest difficulty) from the Open LLM Leaderboard — competition-level math problems requiring multi-step reasoning.
Filters to the hardest MATH subset, preserving headroom among strong models; standardized Open LLM execution settings improve comparability across checkpoints.
Retired by Hugging Face on March 13, 2025; retained scores are historical. Static problems accumulate contamination risk as public solutions spread; competition-style math does not fully represent applied quantitative analysis tasks.
Higher is better. Percentage accuracy on Level 5 problems.
Max score: 100
Retired by Hugging Face on March 13, 2025.
| # | Model | Score |
|---|---|---|
| 1 | 83.1 | |
| 2 | 73.4 | |
| 3 | 72.4 | |
| 4 | 71.9 | |
| 5 | 71.6 | |
| 6 | 71.1 | |
| 7 | 70.5 | |
| 8 | 70.3 | |
| 9 | 69.2 | |
| 10 | 67.3 | |
| 11 | 66.2 | |
| 12 | 62.5 | |
| 13 | 62.4 | |
| 14 | 60.4 | |
| 15 | 60.3 | |
| 16 | 59.8 | |
| 17 | 58.3 | |
| 18 | 58.1 | |
| 19 | 57.2 | |
| 20 | 56.0 | |
| 21 | 55.3 | |
| 22 | 55.2 | |
| 23 | 55.1 | |
| 24 | 54.5 | |
| 25 | 53.9 | |
| 26 | 53.2 | |
| 27 | 53.1 | |
| 28 | 52.8 | |
| 29 | 52.4 | |
| 30 | 52.4 | |
| 31 | 52.3 | |
| 32 | 50.9 | |
| 33 | 50.7 | |
| 34 | 50.1 | |
| 35 | 50.0 | |
| 36 | 49.7 | |
| 37 | 49.6 | |
| 38 | 49.5 | |
| 39 | 49.5 | |
| 40 | 49.5 | |
| 41 | 49.5 | |
| 42 | 49.5 | |
| 43 | 49.3 | |
| 44 | 49.3 | |
| 45 | 49.0 | |
| 46 | 48.3 | |
| 47 | 48.3 | |
| 48 | 48.3 | |
| 49 | 48.3 | |
| 50 | Gemini 2.5 Flash-Lite Preview (Sep '25) (Non-reasoning) source benchpress Evidence observed Apr 13, 2026 | 47.8 |