Vision-Language Compositional Reasoning on Winoground standard (test)
75.5Text ScoreGPT-4o
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| GPT-4oParadigm=Componential Analysis (CA), Reasoning Engine=GPT-4o2025.01 | 75.5 | 58.5 | 52 | |
| Gemini 2.0Paradigm=Componential Analysis (CA), Reasoning Engine=Gemini 2.02025.01 | 71 | 48.75 | 42 | |
| Llama3.3-70BParadigm=Componential Analysis (CA), Reasoning Engine=Llama3.3-70B2025.01 | 68.25 | 49.25 | 41.75 | |
| Qwen2.5-32BParadigm=Componential Analysis (CA), Reasoning Engine=Qwen2.5-32B2025.01 | 67 | 46.25 | 40 | |
| Phi-4-14BParadigm=Componential Analysis (CA), Reasoning Engine=Phi-4-14B2025.01 | 65.25 | 46 | 37.75 | |
| MMICL + CoCoT2025.01 | 64.25 | 52.5 | 50.75 | |
| Qwen2.5-14BParadigm=Componential Analysis (CA), Reasoning Engine=Qwen2.5-14B2025.01 | 59.25 | 34.5 | 27.25 |