Molmo2-O-7B: benchmark results
Provider: Allen AI. Access: Open.
Aggregate Rating 1563 ± 47, rank #1504 of 10916 rated models, from 17 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| BURST-VideoCount | 61.6 | Exact-count accuracy (%; 2.2k queries from BURST test-set tr | 100 |
| Molmo2-CapTest | 40.1 | Caption F1 (%; LLM-judged statement precision and recall aga | 75 |
| Molmo2-VideoCount | 33.2 | Exact-count accuracy (%; 533 validation queries) | 75 |
| PRISM-VLM - Instruction Following | 77.7 | Per-axis mean score (%; mean compliance with 1-3 verifiable | 63.4 |
| Molmo2 Human Preference | 1033 | Elo rating (bootstrap median, Bradley-Terry; pairwise human | 60 |
| Molmo2 Human Preference - Captioning | 1019 | Elo rating (bootstrap median, Bradley-Terry; pairwise human | 60 |
| PRISM-VLM - Visual Discrimination | 81.7 | Per-axis mean score (%; forced A/B choice between the gold c | 50 |
| PRISM-VLM - Multi-Question Robustness | 31.8 | Per-axis mean score (%; share of items whose 3-5 bundled que | 41.5 |
| PRISM-VLM - Quality | 61 | Per-axis mean score (%; base accuracy on the recycled questi | 41.5 |
| EgoIntent - Next Plan | 19.02 | Next-Plan (the action that follows in the video) semantic ag | 35.7 |
| PRISM-VLM - Hallucination Resistance | 39.9 | Per-axis mean score (%; share of trap questions (absent enti | 31.7 |
| PRISM-VLM - Counting | 47.8 | Per-axis mean score (%; exact-integer accuracy on synthesize | 29.3 |
Interactive version: theaggregate.ai/model?slug=molmo2-o-7b · How It Works · Data refreshed daily, snapshot 2026-11-08.