OpenVLM A-Bench - Minor Object Recognition: leaderboard
A-Bench Minor Object Recognition accuracy from the OpenCompass OpenVLM Leaderboard (VLMEvalKit): multiple-choice questions on AI-generated images, identifying small or secondary objects. 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 | InternVL3-8B | 100 | #624 |
| 2 | InternVL3-38B | 100 | #413 |
| 3 | InternVL3-14B | 100 | #506 |
| 4 | GPT-4.1 Mini | 98.3 | #363 |
| 5 | InternVL3-78B | 98.3 | #361 |
| 6 | Qwen 2 VL 72B | 98.3 | #450 |
| 7 | Gemini 1.5 Pro | 98.3 | #454 |
| 8 | Aquila-VL-2B | 98.3 | #1216 |
| 9 | Qwen 2.5 VL 7B | 98.3 | #641 |
| 10 | Gemini 1.5 Flash | 98.3 | #620 |
| 11 | Grok 2 (1212) | 98.3 | #502 |
| 12 | GPT-4.1 Nano | 96.6 | #730 |
| 13 | InternVL2-8B | 96.6 | #813 |
| 14 | MiniCPM-V-2.6 | 96.6 | #827 |
| 15 | Qwen 2 VL 7B | 96.6 | #743 |
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-minor-object-recognition · How It Works · Data refreshed daily, snapshot 2026-11-05.