Fine-grained Provenance Generation on PROVE-ASQA
4.36Text QualityGemini 2.5 Pro
Evaluation Results
| Method | Links | ||||||
|---|---|---|---|---|---|---|---|
| Gemini 2.5 ProModel Type=Closed-Source2026.01 | 4.36 | 3.23 | 3.95 | 1.28 | 0.16 | 2.6 | |
| GenProveModel Type=Ours2026.01 | 4.29 | 3.83 | 4.16 | 2.57 | 1.94 | 3.36 | |
| GPT-5Model Type=Closed-Source2026.01 | 4.24 | 3.17 | 3.22 | 2.33 | 0.63 | 2.72 | |
| KimiModel Type=Closed-Source2026.01 | 4.21 | 2.44 | 3.85 | 0.38 | 0.21 | 2.22 | |
| GLM-4.5-355BModel Type=Open-Source2026.01 | 4.15 | 2.9 | 3.59 | 0.99 | 0.3 | 2.39 | |
| Qwen3-8BModel Type=Open-Source2026.01 | 4.03 | 2.63 | 3.36 | 0.82 | 1.26 | 2.42 | |
| GLM-4-9BModel Type=Open-Source2026.01 | 4.02 | 2.45 | 2.85 | 0.33 | 0.92 | 2.11 | |
| Qwen3-14BModel Type=Open-Source2026.01 | 3.99 | 2.91 | 3.25 | 1.2 | 0.47 | 2.36 | |
| Yi-1.5-9BModel Type=Open-Source2026.01 | 3.98 | 2.05 | 2.44 | 0.1 | 0.2 | 1.76 | |
| InternLM2.5-7BModel Type=Open-Source2026.01 | 3.91 | 1.49 | 1.51 | 1.16 | 0.1 | 1.64 | |
| Gemma-3-12BModel Type=Open-Source2026.01 | 3.9 | 3.03 | 3.69 | 1.14 | 0.96 | 2.54 | |
| Llama-3.1-8BModel Type=Open-Source2026.01 | 3.82 | 2.4 | 2.74 | 1.16 | 1.2 | 2.27 | |
| Baichuan2-7BModel Type=Open-Source2026.01 | 3.65 | 0.41 | 0.51 | 0.2 | 0.1 | 0.98 | |
| Hunyuan-7BModel Type=Open-Source2026.01 | 3.45 | 2.38 | 3.12 | 0.61 | 1.08 | 2.13 | |
| Vicuna-7b-v1.5Model Type=Open-Source2026.01 | 2.82 | 0.62 | 0.75 | 0.16 | 0.41 | 0.95 |