OpenVLM A-Bench - Attribute Awareness: leaderboard
A-Bench Attribute Awareness accuracy from the OpenCompass OpenVLM Leaderboard (VLMEvalKit): multiple-choice questions on AI-generated images, checking whether object attributes such as colour or material match the image. A dimension of A-Bench-P1 (high-level semantic understanding). 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 | InternVL3-78B | 96.4 | #361 |
| 2 | Qwen 2 VL 72B | 96.4 | #450 |
| 3 | InternVL3-38B | 95.5 | #413 |
| 4 | Gemini 1.5 Pro | 95.5 | #454 |
| 5 | GPT-4.1 (2025-04-14) | 93.6 | #257 |
| 6 | InternVL3-8B | 92.7 | #624 |
| 7 | Pixtral-12B | 92.7 | #805 |
| 8 | Qwen 2 VL 7B | 91.8 | #743 |
| 9 | Grok 2 (1212) | 91.8 | #502 |
| 10 | GPT-4.1 Mini | 90.9 | #363 |
| 11 | MiniCPM-V-2.6 | 90.9 | #827 |
| 12 | Gemma 3 27B | 90.9 | #606 |
| 13 | Qwen 2 VL 2B | 90.9 | #1191 |
| 14 | InternVL3-14B | 90 | #506 |
| 15 | Claude 3.7 Sonnet | 90 | #259 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=openvlm-a-bench-attribute-awareness · How It Works · Data refreshed daily, snapshot 2026-11-05.