OODBench (OOD-Hard): leaderboard
Metric: Accuracy (%) on balanced yes/no existence questions ('Does this image contain / not contain a [class]?', opposite labels, so random choice scores 50) about OOD-Hard (flagged as out-of-distribution by both scoring models, CLIP and BLIP-2) instance-category pairs from COCO, LVIS, nuScenes and Cityscapes; higher is better. Source: arxiv.org. Saturation forecast: Estimated already saturated. 10 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Qwen 2.5 VL 7B Instruct | 68.7 | #643 |
| 2 | Qwen 2 VL 7B Instruct | 66.77 | #816 |
| 3 | GPT-4o | 65.13 | #333 |
| 4 | InternVL2-8B | 59.05 | #826 |
No result here: #3 Claude Opus 5.5, #5 GPT-6 Astra, #8 Claude Fable 5.1.
Interactive version: theaggregate.ai/benchmark?slug=oodbench-ood-hard · How It Works · Data refreshed daily, snapshot 2026-10-11.