GPQA (Graduate-level science) from the Open LLM Leaderboard — PhD-difficulty questions in physics, chemistry, and biology.
Runs GPQA with consistent leaderboard settings, enabling cleaner open-model comparisons; difficult science questions reduce gains from shallow pattern-matching heuristics.
Retired by Hugging Face on March 13, 2025; retained scores are historical. Domain coverage is limited to three sciences and may not transfer to broader reasoning workloads; overlap with other public GPQA leaderboards increases potential benchmark familiarity.
Higher is better. Percentage accuracy.
Max score: 100
Retired by Hugging Face on March 13, 2025.
| # | Model | Score |
|---|---|---|
| 1 | 43.7 | |
| 2 | 40.7 | |
| 3 | 39.3 | |
| 4 | 38.9 | |
| 5 | 38.8 | |
| 6 | 38.8 | |
| 7 | 38.8 | |
| 8 | 38.6 | |
| 9 | 38.2 | |
| 10 | 38.2 | |
| 11 | 37.9 | |
| 12 | 37.5 | |
| 13 | 37.5 | |
| 14 | 37.4 | |
| 15 | 37.3 | |
| 16 | 37.2 | |
| 17 | 37.2 | |
| 18 | 37.0 | |
| 19 | DeepHermes 3 - Mistral 24B Preview (Non-reasoning) source open_llm_leaderboard Evidence observed Sep 19, 2026 | 37.0 |
| 20 | 37.0 | |
| 21 | 36.9 | |
| 22 | 36.7 | |
| 23 | 36.6 | |
| 24 | 36.6 | |
| 25 | 36.6 | |
| 26 | 36.4 | |
| 27 | Gemini 2.5 Flash-Lite Preview (Sep '25) (Non-reasoning) source benchpress Evidence observed Apr 13, 2026 | 36.2 |
| 28 | 36.2 | |
| 29 | 36.2 | |
| 30 | 36.1 | |
| 31 | 36.0 | |
| 32 | 35.9 | |
| 33 | 35.8 | |
| 34 | 35.8 | |
| 35 | 35.8 | |
| 36 | 35.8 | |
| 37 | 35.8 | |
| 38 | 35.7 | |
| 39 | 35.7 | |
| 40 | 35.7 | |
| 41 | 35.7 | |
| 42 | 35.6 | |
| 43 | 35.5 | |
| 44 | 35.5 | |
| 45 | 35.3 | |
| 46 | 35.3 | |
| 47 | 35.2 | |
| 48 | 35.1 | |
| 49 | 35.0 | |
| 50 | 34.9 |