Molmo2 Human Preference: leaderboard
Pairwise human preference study from the Molmo2 paper (Allen Institute for AI and University of Washington, 2026): annotators compare two models' answers to 450 open-ended questions about videos and their captions of 51 videos, over 105,000 ratings in all. Scores are Bradley-Terry Elo ratings (bootstrap median over 1,000 rounds) relative to the models in the study; higher is better.
Metric: Elo rating (bootstrap median, Bradley-Terry; pairwise human preferences on 450 video questions and 51 captioning videos). Source: arxiv.org. Saturation forecast: Around March 2027. 21 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Gemini 2.5 Pro | 1096 | #262 |
| 2 | Gemini 2.5 Flash | 1084 | #452 |
| 3 | Gemini 3 Pro | 1082 | #135 |
| 4 | GPT-5 Mini | 1076 | #349 |
| 5 | Molmo2-8B | 1057 | #1488 |
| 6 | GPT-5 | 1031 | #163 |
| 7 | Claude Sonnet 4.5 | 1008 | #244 |
| 8 | GLM-4.1V-9B (Thinking) | 962 | #1061 (GLM-4.1V-9B) |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=molmo2-human-preference · How It Works · Data refreshed daily, snapshot 2026-11-12.