Visual Question Answering on VCR
68.68Overall AccuracyTTH
Evaluation results
| Method | Links | ||
|---|---|---|---|
| TTHTarget Model=GPT-5 Nano, Zero-shot transfer=true2026.05 | 68.68 | 24.22 | |
| External judgeTarget Model=GPT-5 Nano, Zero-shot transfer=true2026.05 | 68.64 | 21.62 | |
| TTHTarget Model=Gemini 2.5 Flash Lite, Zero-shot transfer=true2026.05 | 67.04 | 23.18 | |
| Self refineTarget Model=Gemini 2.5 Flash Lite, Zero-shot transfer=true2026.05 | 65.4 | 14.58 | |
| CoTTarget Model=GPT-5 Nano, Zero-shot transfer=true2026.05 | 64.04 | 18.82 | |
| External judgeTarget Model=Gemini 2.5 Flash Lite, Zero-shot transfer=true2026.05 | 63.66 | 19.27 | |
| TTHTarget Model=Claude 4.5 Haiku, Zero-shot transfer=true2026.05 | 63.44 | 34.43 | |
| Self refineTarget Model=GPT-5 Nano, Zero-shot transfer=true2026.05 | 63.2 | 11.89 | |
| CoTTarget Model=Gemini 2.5 Flash Lite, Zero-shot transfer=true2026.05 | 63.02 | 14.58 | |
| BaseTarget Model=GPT-5 Nano, Zero-shot transfer=true2026.05 | 63 | — | |
| External judgeTarget Model=Claude 4.5 Haiku, Zero-shot transfer=true2026.05 | 62.6 | 21.35 | |
| Self refineTarget Model=Claude 4.5 Haiku, Zero-shot transfer=true2026.05 | 62.2 | 11.46 | |
| CoTTarget Model=Claude 4.5 Haiku, Zero-shot transfer=true2026.05 | 62.06 | 15.1 | |
| BaseTarget Model=Gemini 2.5 Flash Lite, Zero-shot transfer=true2026.05 | 61.68 | — | |
| BaseTarget Model=Claude 4.5 Haiku, Zero-shot transfer=true2026.05 | 61.64 | — |