Open LLM Leaderboard - GPQA: leaderboard
The "GPQA" subtask of Open LLM Leaderboard. A task from the Open LLM Leaderboard, the standard open-weight evaluation suite spanning knowledge, reasoning, math and truthfulness benchmarks.
Metric: Score. Source: huggingface.co. Saturation forecast: Estimated already saturated. 4576 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Mistral Large 2 (Nov) Instruct (2411) | 43.71 | #465 |
| 2 | Qwen 2.5 32B | 41.19 | #546 |
| 3 | Phi-4 | 40.6 | #706 |
| 4 | Qwen 2.5 72B | 40.52 | #499 |
| 5 | phi-4-14B | 40.35 | #733 |
| 6 | calme-3.2-instruct-78B | 40.27 | #347 |
| 7 | Llama 3 70B | 39.77 | #631 |
| 8 | Rombos-LLM-V2.5-Qwen-32B | 39.68 | #367 |
| 9 | Saka-14B | 39.6 | #344 |
| 10 | Qwen 2 72B | 39.43 | #521 |
| 11 | Llama 3.1 70B | 38.76 | #592 |
| 12 | DeepSeek R1 Distill Qwen 14B | 38.76 | #844 |
| 13 | Mistral-Small-24B-Base-2501 | 38.76 | #670 |
| 14 | Qwentile2.5-32B-Instruct | 38.42 | #339 |
| 15 | oxyge1-33B | 38.26 | #690 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=open-llm-leaderboard-gpqa · How It Works · Data refreshed daily, snapshot 2026-10-20.