Safety
HallusionBench
Build HallusionBench per-item response matrix for the shipped GPT-4V reference run
200items
1subjects
BSD-3-Clauselicense
safetydomain
reasoningdomain
textmodality
imagemodality
item-level responses released
Saturation status: Unknown
Response matrix
Fit to width. Hover for subject & item; click a cell for details.
Correct (1)Incorrect (0)Unobserved
Scale: 1 = correct · 0 = incorrect