Multimodal Hallucination Assessment on HallusionBench
70AccuracyQwen3.5-27B
Evaluation Results
| Method | Links | |
|---|---|---|
| Qwen3.5-27BMode=REASONING, Architecture=Dense, # Total Params=27B, # Activated Params=27B2026.04 | 70 | |
| Qwen3-VL-32BMode=Thinking, Architecture=Dense, # Total Params=33B, # Activated Params=33B2026.04 | 67.4 | |
| Qwen3-VL-235B-A22BMode=Thinking, Architecture=MoE, # Total Params=236B, # Activated Params=23B2026.04 | 66.7 | |
| EXAONE 4.5 33BMode=REASONING, Architecture=Dense, # Total Params=33B, # Activated Params=33B2026.04 | 63.7 | |
| GPT-5 miniMode=REASONING: HIGH, Architecture=-, # Total Params=-, # Activated Params=-2026.04 | 63.2 | |
| F3ARatio=60%, Backbone=InternVL3.5-8B2026.05 | 54.88 | |
| InternVL3.5-8BRatio=100%, Backbone=InternVL3.5-8B2026.05 | 54.77 | |
| VisionZipRatio=60%, Backbone=InternVL3.5-8B2026.05 | 54.22 | |
| FastVRatio=60%, Backbone=InternVL3.5-8B2026.05 | 53.35 | |
| DivPruneRatio=60%, Backbone=InternVL3.5-8B2026.05 | 53.35 | |
| DivPruneRatio=40%, Backbone=InternVL3.5-8B2026.05 | 52.42 | |
| CDPrunerRatio=60%, Backbone=InternVL3.5-8B2026.05 | 52.35 | |
| VisionZipRatio=40%, Backbone=InternVL3.5-8B2026.05 | 52.25 | |
| F3ARatio=40%, Backbone=InternVL3.5-8B2026.05 | 52.25 | |
| CDPrunerRatio=40%, Backbone=InternVL3.5-8B2026.05 | 51.83 | |
| FastVRatio=40%, Backbone=InternVL3.5-8B2026.05 | 51.76 | |
| F3ARatio=20%, Backbone=InternVL3.5-8B2026.05 | 50.55 | |
| CDPrunerRatio=20%, Backbone=InternVL3.5-8B2026.05 | 48.46 | |
| DivPruneRatio=20%, Backbone=InternVL3.5-8B2026.05 | 48.42 | |
| VisionZipRatio=20%, Backbone=InternVL3.5-8B2026.05 | 48.31 | |
| FastVRatio=20%, Backbone=InternVL3.5-8B2026.05 | 47.27 |