General
Vript-Hard
Vript-Hard: video-understanding challenge tasks on densely-captioned clips. Vript-RR asks scene-specific multiple-choice questions given a hint; Vript-ERO asks for the correct temporal order of three scenes. Responses are per-(video-LLM, item) graded outcomes reproduced from the upstream verification pipeline.
286items
3subjects
Apache-2.0license
reasoningdomain
videomodality
textmodality
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
Subjects
- 1VideoChat20.4704
- 2VideoChat0.2796
- 3Claude-Opus0.2313