List your company

Multimodal Fact-Level Attribution on Video-MMMU 1.0 (sampled examples)

86.8AccuracyGemini-3-Flash
38.12850.76463.476.036Feb 12, 2026

Evaluation results

MethodLinks
Gemini-3-FlashVariant=Base2026.0286.8———
Gemini-3-FlashVariant=+ Post-Hoc Attribution2026.0286.887.947.244.1
Gemini-3-FlashVariant=+ Citation2026.028688.264.556.9
Gemini-3-ProVariant=+ Citation2026.028663.467.341.8
Gemini-3-ProVariant=Base2026.0285.3———
Gemini-3-ProVariant=+ Post-Hoc Attribution2026.0285.36843.736.9
Gemini-2.5-FlashVariant=+ Citation2026.0284.96363.441.5
Gemini-2.5-FlashVariant=Base2026.0284.2———
Gemini-2.5-FlashVariant=+ Post-Hoc Attribution2026.0284.273.844.938
Qwen-3-VL-ThinkingCategory=Vision-Language Only, Variant=+ Citation2026.026023.215.17.6
Qwen-3-VL-InstructCategory=Vision-Language Only, Variant=+ Citation2026.025530.240.117.5
Qwen-Omni-ThinkingVariant=Base2026.0253———
Qwen-Omni-ThinkingVariant=+ Post-Hoc Attribution2026.025376.316.812.8
Qwen-3-VL-InstructCategory=Vision-Language Only, Variant=Base2026.0253———
Qwen-3-VL-InstructCategory=Vision-Language Only, Variant=+ Post-Hoc Attribution2026.025393.444.642.3
Qwen-Omni-ThinkingVariant=+ Citation2026.025136.37.64.8
Qwen-3-VL-ThinkingCategory=Vision-Language Only, Variant=Base2026.0251———
Qwen-3-VL-ThinkingCategory=Vision-Language Only, Variant=+ Post-Hoc Attribution2026.025154.331.518.9
Molmo2Category=Vision-Language Only, Variant=Base2026.0250.5———
Molmo2Category=Vision-Language Only, Variant=+ Post-Hoc Attribution2026.0250.566.41511.4
Qwen-Omni-InstructVariant=Base2026.0245———
Qwen-Omni-InstructVariant=+ Post-Hoc Attribution2026.024595.117.917.6
Molmo2Category=Vision-Language Only, Variant=+ Citation2026.0244.382.621.419.3
Qwen-Omni-InstructVariant=+ Citation2026.024034.621.89.8
Gemini-2.5-Flash2026.02—57.132.521.8
Gemini-3-Pro2026.02—59.724.816.3
Qwen3-Omni-Instruct2026.02—35.914.95.6
Qwen3-Omni-Thinking2026.02—4523.421.8