OpenVLM A-Bench: leaderboard
OpenCompass OpenVLM evaluation of A-Bench: attribute and compositional image understanding with questions about object attributes, spatial arrangement, occlusion, orientation, size, and relations.
Metric: Accuracy (%). Source: huggingface.co. Saturation forecast: Estimated already saturated. 160 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-4.1 (2025-04-14) | 79.6 | #250 |
| 2 | Qwen 2 VL 72B | 79.4 | #460 |
| 3 | InternVL3-78B | 78.9 | #352 |
| 4 | GPT-4.1 Mini | 77.4 | #355 |
| 5 | InternVL3-38B | 77.1 | #405 |
| 6 | Qwen 2 VL 7B | 76.4 | #750 |
| 7 | InternVL3-8B | 76.3 | #620 |
| 8 | InternVL3-14B | 76.3 | #505 |
| 9 | Gemma 3 27B | 76.3 | #607 |
| 10 | Grok 2 (1212) | 76.3 | #496 |
| 11 | Qwen 2.5 VL 7B | 75.7 | #643 |
| 12 | Gemini 1.5 Pro | 75.4 | #445 |
| 13 | Aquila-VL-2B | 75.4 | #1238 |
| 14 | MiniCPM-V-2.6 | 74.6 | #836 |
| 15 | Moonshot v1 8K | 74.3 | #617 |
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 · How It Works · Data refreshed daily, snapshot 2026-10-20.