Global PIQA - Non-Parallel (Strict Exact Match)
General QA · 2025-10-28
Global PIQA's non-parallel split contains locally authored physical-commonsense questions with two answer options. This definition scores generated answers using exact match with strict answer extraction.
Top models (higher is better)
| Model | Score |
|---|---|
| Gemma 4 31B IT | 84.9 |
| Gemma 3 27B IT | 81.7 |
| Command A | 81.2 |
| Qwen3.5-27B | 78.5 |
| Qwen2.5 72B Instruct | 78.3 |
| Gemma 3 12B IT | 78.1 |
| gpt-oss-20b | 77.8 |
| Qwen3 14B | 77.1 |
| Llama 3.1 70B Instruct | 76.2 |
| Qwen3 32B | 75.9 |
| Qwen2.5 Instruct 32B | 75.2 |
| Apertus 70B Instruct | 74.3 |
| Phi-4 | 74.2 |
| Qwen3 8B | 74.0 |
| Qwen3.5-9B | 73.5 |
| Gemma 2 27B IT | 73.4 |
| Qwen2.5 14B Instruct | 73.4 |
| DeepSeek-R1-Distill-Qwen-32B | 72.2 |
| Aya Expanse 32B | 70.9 |
| DeepSeek R1 Distill Qwen 14B | 70.7 |
| Qwen3 4B | 70.6 |
| gpt-oss-120b | 69.5 |
| C4AI Command R 08-2024 | 68.9 |
| Gemma 2 9B IT | 68.3 |
| Gemma 3 4B IT | 67.6 |
| Apertus 8B Instruct | 67.0 |
| Qwen2.5 7B Instruct | 66.5 |
| C4AI Command R+ 08-2024 | 65.8 |
| Llama 3.1 8B Instruct | 61.0 |
| Qwen3 1.7B | 60.8 |
| Qwen2.5 3B Instruct | 60.2 |
| Phi-3 Medium 4K Instruct | 59.5 |
| Tiny Aya Global | 59.5 |
| Command R7B (Dec 2024) | 59.0 |
| Phi-4-mini-instruct | 58.5 |
| Gemma 2 2B IT | 58.4 |
| Phi-3.5 Mini Instruct | 56.9 |
| Phi-3-mini-4k-instruct | 55.2 |
| Gemma 3 1B IT | 54.6 |
| Mistral 7B Instruct v0.3 | 54.6 |