Human Preference & Alignment
InterMT
52.6k multi-turn multimodal dialogues with 32.4k human-labeled preference pairs.
12,557items
2subjects
100%observed
Modelsubject type
Response matrix
Every model, scored item by item.
Each row is an AI model and each column an item, ordered so the strongest models and easiest items gather toward one corner. 2 subjects × 12,557 items, 100% of cells evaluated.
Fit to width. Hover for subject & item; click a cell for details.
lowhighUnobserved
Scale: 0 to 3 human rating (per dimension): nine multimodal-dialogue quality dimensions (helpfulness, coherence, image–text consistency, …); higher is better.
Sample items
What the questions look like — and how subjects answer.
A spread of items across the difficulty range, each shown with a few subjects' actual answers.
Subjects
The models, agents, and reward models evaluated.
2 subjects, ranked by mean response across this benchmark's items.
- 14a8db3fc2.2
- 22b13d0f82.178