Multimodal Fact-Level Attribution on WorldSense 1.0 (sampled examples)
71.4AccuracyGemini-3-Pro
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| Gemini-3-ProVariant=Base2026.02 | 71.4 | — | — | — | |
| Gemini-3-ProVariant=+ Post-Hoc Attribution2026.02 | 71.4 | 97 | 67.1 | 65.2 | |
| Gemini-3-ProVariant=+ Citation2026.02 | 70 | 78.3 | 64.9 | 51.7 | |
| Gemini-3-FlashVariant=Base2026.02 | 67 | — | — | — | |
| Gemini-3-FlashVariant=+ Post-Hoc Attribution2026.02 | 67 | 95.1 | 71.4 | 69.2 | |
| Gemini-2.5-FlashVariant=+ Citation2026.02 | 66.5 | 81.2 | 65.4 | 54.1 | |
| Gemini-3-FlashVariant=+ Citation2026.02 | 66.2 | 95.9 | 66.5 | 64.4 | |
| Gemini-2.5-FlashVariant=Base2026.02 | 62.3 | — | — | — | |
| Gemini-2.5-FlashVariant=+ Post-Hoc Attribution2026.02 | 62.3 | 97.4 | 62.3 | 60.8 | |
| Qwen-Omni-ThinkingVariant=+ Citation2026.02 | 61 | 52.7 | 56.3 | 31.3 | |
| Qwen-Omni-InstructVariant=Base2026.02 | 57 | — | — | — | |
| Qwen-Omni-InstructVariant=+ Post-Hoc Attribution2026.02 | 57 | 99.5 | 45.7 | 45.4 | |
| Qwen-Omni-ThinkingVariant=Base2026.02 | 56.5 | — | — | — | |
| Qwen-Omni-ThinkingVariant=+ Post-Hoc Attribution2026.02 | 56.5 | 93.2 | 60 | 56.3 | |
| Qwen-Omni-InstructVariant=+ Citation2026.02 | 54 | 47.6 | 53.3 | 29 | |
| Qwen-3-VL-InstructCategory=Vision-Language Only, Variant=Base2026.02 | 50 | — | — | — | |
| Qwen-3-VL-InstructCategory=Vision-Language Only, Variant=+ Post-Hoc Attribution2026.02 | 50 | 98.9 | 70.2 | 69.4 | |
| Qwen-3-VL-ThinkingCategory=Vision-Language Only, Variant=+ Citation2026.02 | 49 | 38.5 | 56.1 | 30.8 | |
| Qwen-3-VL-InstructCategory=Vision-Language Only, Variant=+ Citation2026.02 | 48 | 39 | 52 | 25.5 | |
| Qwen-3-VL-ThinkingCategory=Vision-Language Only, Variant=Base2026.02 | 47 | — | — | — | |
| Qwen-3-VL-ThinkingCategory=Vision-Language Only, Variant=+ Post-Hoc Attribution2026.02 | 47 | 76.6 | 58.9 | 48.2 | |
| Molmo2Category=Vision-Language Only, Variant=Base2026.02 | 41 | — | — | — | |
| Molmo2Category=Vision-Language Only, Variant=+ Post-Hoc Attribution2026.02 | 41 | 75 | 38.3 | 33.2 | |
| Molmo2Category=Vision-Language Only, Variant=+ Citation2026.02 | 40 | 69.1 | 50.2 | 39.7 | |
| Gemini-2.5-Flash2026.02 | — | 85 | 65.8 | 59.9 | |
| Gemini-3-Pro2026.02 | — | 76.8 | 61.6 | 49.7 | |
| Qwen3-Omni-Instruct2026.02 | — | 55.1 | 35.4 | 27.3 | |
| Qwen3-Omni-Thinking2026.02 | — | 47.1 | 41.2 | 23.1 |