Fine-grained Hate Categorization on H-VLI (Hard)
78.79AccuracyGemini-2.5-Pro
Evaluation results
| Method | Links | |
|---|---|---|
| Gemini-2.5-ProTraining=False, Framework=ARCADE2026.03 | 78.79 | |
| GPT-4oTraining=False, Framework=ARCADE2026.03 | 75.76 | |
| Gemini-2.5-ProTraining=False2026.03 | 74.56 | |
| Gemini-2.5-FlashTraining=False, Framework=ARCADE2026.03 | 74.55 | |
| Gemini-2.5-FlashTraining=False2026.03 | 69.23 | |
| GPT-5Training=False2026.03 | 69.23 | |
| GPT-4.1-miniTraining=False2026.03 | 66.27 | |
| GPT-5Training=False, Framework=ARCADE2026.03 | 65.54 | |
| GPT-4oTraining=False2026.03 | 64.5 | |
| GPT-5-miniTraining=False, Framework=ARCADE2026.03 | 63.64 | |
| GPT-4.1-miniTraining=False, Framework=ARCADE2026.03 | 58.18 | |
| Qwen-VL-MaxTraining=False2026.03 | 55.15 | |
| Qwen3-VL-PlusTraining=False, Framework=ARCADE2026.03 | 51.52 | |
| GPT-5-miniTraining=False2026.03 | 49.11 | |
| Qwen-VL-PlusTraining=False2026.03 | 44.24 | |
| Qwen3-VL-PlusTraining=False2026.03 | 40 | |
| Qwen-VL-MaxTraining=False, Framework=ARCADE2026.03 | 39.39 | |
| ViT-b-16Training=True2026.03 | 37.28 | |
| PromptHateTraining=True2026.03 | 36.69 | |
| BERT+ViTTraining=True2026.03 | 33.73 | |
| MMBTTraining=True2026.03 | 33.14 | |
| MomentaTraining=True2026.03 | 30.18 | |
| BERTTraining=True2026.03 | 29.59 | |
| Qwen-VL-PlusTraining=False, Framework=ARCADE2026.03 | 28.48 |