DO-Bench - Prior Robustness: leaderboard
Resistance to misleading context in object-existence questions on DO-Bench (2026): across four levels of textual prior pressure, the model must keep affirming a present but unusual object and keep denying an expected object that was removed from the image. PriorRobust in percent (any constant answer scores 50); higher is better.
Metric: PriorRobust (%): one minus the normalized area under the false-negative curve (present object, prior strength 0-3) and under the false-positive curve (absent object, prior strength 0-3), averaged; any constant answer scores 50, on DO-Bench, 124 scenes with ten controlled yes/no object-existence queries each (1,240 queries: a present-but-anomalous object under four prior strengths and two zoomed views, an absent-but-expected object removed by local inpainting under four prior strengths), expert-verified images, temperature 0 or the mean of three API runs; higher is better. Source: arxiv.org. Saturation forecast: Estimated already saturated. 17 models tracked. Updated 7 Oct 2026.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Gemini 3 Flash (Preview) | 91.45 | #156 |
| 2 | GPT-5.2 | 84.64 | #190 |
| 3 | Claude Opus 4.6 | 81.45 | #114 |
| 4 | InternVL2.5-8B | 65.72 | #2393 |
| 5 | InternVL2.5-78B | 63.78 | #1036 |
| 6 | Qwen 2.5 VL 32B Instruct | 63.31 | #1065 |
| 7 | Qwen 2.5 VL 72B Instruct | 61.62 | #849 |
| 8 | InternVL2.5-38B | 58.87 | #1383 |
| 9 | Qwen 2.5 VL 7B Instruct | 56.45 | #1919 |
| 10 | Qwen 2.5 VL 3B Instruct | 53.36 | #2720 |
| 11 | InternVL2.5-4B | 49.8 | #2573 |
| 12 | InternVL2.5-2B | 36.7 | #6523 |
| 13 | LLaVA-v1.5-7B | 21.78 | #7642 |
| 14 | Qwen-VL-7B | 21.25 | |
| 15 | InstructBLIP-Vicuna-13B | 14.04 | #8813 |
No result here: #2 Claude Fable 5.2, #3 Gemini 4 Neon, #4 GPT-6.2 Astra.
Interactive version: theaggregate.ai/benchmark?slug=do-bench-prior-robustness · How It Works · Data refreshed daily, snapshot 2026-11-22.