OpenVLM Q-Bench - Yes-or-No (Distortion): leaderboard
Q-Bench accuracy on yes-or-no questions about image-wide distortions such as blur, noise or exposure, from the OpenCompass OpenVLM Leaderboard (VLMEvalKit) on the Q-Bench dev split. Q-Bench probes low-level visual perception; each of its twelve question-type by concern cells holds 70-188 of the 1,495 dev questions. Accuracy in percent, higher is better.
Metric: Accuracy (%). Source: huggingface.co. Saturation forecast: Estimated already saturated. 160 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Qwen 2 VL 72B | 91 | #450 |
| 2 | Qwen 2 VL 7B | 89.4 | #743 |
| 3 | GPT-4.1 Mini | 88.8 | #363 |
| 4 | GPT-4.1 (2025-04-14) | 86.2 | #257 |
| 5 | MiniCPM-V-2.6 | 86.2 | #827 |
| 6 | InternVL3-78B | 85.6 | #361 |
| 7 | Qwen 2 VL 2B | 85.6 | #1191 |
| 8 | Moonshot v1 8K | 85.1 | #622 |
| 9 | InternVL3-38B | 84 | #413 |
| 10 | InternVL3-8B | 83.5 | #624 |
| 11 | Qwen 2.5 VL 7B | 83.5 | #641 |
| 12 | Aquila-VL-2B | 82.4 | #1216 |
| 13 | Gemini 1.5 Pro | 81.9 | #454 |
| 14 | GPT-4.1 Nano | 81.4 | #730 |
| 15 | InternVL3-14B | 81.4 | #506 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=openvlm-q-bench-yes-or-no-distortion · How It Works · Data refreshed daily, snapshot 2026-11-05.