Misinformation Detection on MISBENCH (Multi-hop based Misinformation) 1.0 (test)
96.88Factual Memory Success RateGPT-4o
Evaluation results
| Method | Links | |||||||||
|---|---|---|---|---|---|---|---|---|---|---|
| GPT-4oModel Series=Closed-source Models2025.05 | 96.88 | 93.81 | 3.07 | 98.28 | 97.68 | 0.6 | 96.57 | 94.33 | 2.24 | |
| Qwen2.5-14BModel Series=Qwen2.5 Series2025.05 | 79.08 | 68.98 | 10.1 | 99.63 | 99.43 | 0.2 | 73.66 | 68.86 | 4.8 | |
| Qwen2.5-72BModel Series=Qwen2.5 Series2025.05 | 75.96 | 58.55 | 17.41 | 90.15 | 81.86 | 8.29 | 67.56 | 52.8 | 14.76 | |
| Claude3.5-haikuModel Series=Closed-source Models2025.05 | 71.43 | 61.71 | 9.72 | 87.14 | 87.04 | 0.1 | 66.86 | 62.74 | 4.12 | |
| LLaMA3-70BModel Series=LLaMA3 Series2025.05 | 70.32 | 58.82 | 11.5 | 91.47 | 84.8 | 6.67 | 69.49 | 64.57 | 4.92 | |
| Qwen2.5-3BModel Series=Qwen2.5 Series2025.05 | 64.02 | 57.88 | 6.14 | 88.2 | 86.76 | 1.44 | 59.36 | 52.34 | 7.02 | |
| Gemma2-9BModel Series=Gemma2 Series2025.05 | 58.93 | 50.51 | 8.42 | 92.94 | 90.63 | 2.31 | 52.07 | 48.38 | 3.69 | |
| Gemma2-2BModel Series=Gemma2 Series2025.05 | 56.97 | 51.58 | 5.39 | 84.74 | 81.31 | 3.43 | 52.9 | 50.18 | 2.72 | |
| Gemma2-27BModel Series=Gemma2 Series2025.05 | 46.55 | 32.36 | 14.19 | 79.39 | 70.4 | 8.99 | 37.84 | 29.08 | 8.76 | |
| DeepSeek-V2.5Model Series=Closed-source Models2025.05 | 46.39 | 38.11 | 8.28 | 69.31 | 68.21 | 1.1 | 41.52 | 34.95 | 6.57 | |
| Qwen2.5-7BModel Series=Qwen2.5 Series2025.05 | 21.75 | 15.73 | 6.02 | 55.14 | 52.5 | 2.64 | 18.28 | 13.16 | 5.12 | |
| LLaMA3-8BModel Series=LLaMA3 Series2025.05 | 20.02 | 17.29 | 2.73 | 49.94 | 46.78 | 3.16 | 23.43 | 18.35 | 5.08 |