Multi-hop Question Answering on 2WikiMHQA (test)
71.85EMGemini-2.5-Flash
Evaluation results
| Method | Links | ||
|---|---|---|---|
| Gemini-2.5-FlashType=API2026.02 | 71.85 | 80.59 | |
| Deepseek-V3.2Type=API2026.02 | 67.45 | 76.66 | |
| CRAFT7BType=Trained2026.02 | 67.15 | 73.77 | |
| GPT-5-miniType=API2026.02 | 65.28 | 76.53 | |
| Qwen3-30BType=Pre-Trained2026.02 | 64.7 | 73.71 | |
| Qwen3-4BType=Pre-Trained2026.02 | 57.7 | 67.43 | |
| Qwen2.5-7BType=Trained2026.02 | 56.15 | 63.79 | |
| Qwen2.5-7BType=Pre-Trained2026.02 | 50.6 | 59.29 | |
| Qwen2.5-3BType=Trained2026.02 | 50.4 | 57.75 | |
| CRAFT3BType=Trained2026.02 | 49.3 | 57.19 | |
| Qwen2.5-3BType=Pre-Trained2026.02 | 41.45 | 50.09 | |
| CRAFT1.5BType=Trained2026.02 | 38.4 | 46.19 | |
| Qwen2.5-1.5BType=Trained2026.02 | 34.3 | 40.92 | |
| Qwen2.5-1.5BType=Pre-Trained2026.02 | 26.05 | 33.29 | |
| Qwen2.5-0.5BType=Pre-Trained2026.02 | 6.4 | 12.37 | |
| Qwen2.5-0.5BType=Trained2026.02 | 2.8 | 4.7 | |
| CRAFT0.5BType=Trained2026.02 | 0.05 | 0.05 |