OpenVLM A-Bench - Major Object Recognition: leaderboard
A-Bench Major Object Recognition accuracy from the OpenCompass OpenVLM Leaderboard (VLMEvalKit): multiple-choice questions on AI-generated images, identifying the main objects in 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 | GPT-4.1 Mini | 95.6 | #363 |
| 2 | InternVL2-8B | 95.6 | #813 |
| 3 | GPT-4.1 (2025-04-14) | 94.5 | #257 |
| 4 | Qwen 2 VL 72B | 94.5 | #450 |
| 5 | Qwen 2.5 VL 7B | 94.5 | #641 |
| 6 | Moonshot v1 8K | 94.5 | #622 |
| 7 | MiniCPM-V-2.6 | 94 | #827 |
| 8 | Qwen 2 VL 7B | 94 | #743 |
| 9 | InternVL3-8B | 93.4 | #624 |
| 10 | InternVL3-78B | 93.4 | #361 |
| 11 | InternVL3-14B | 93.4 | #506 |
| 12 | Aquila-VL-2B | 93.4 | #1216 |
| 13 | InternVL3-38B | 92.9 | #413 |
| 14 | Qwen 2 VL 2B | 92.9 | #1191 |
| 15 | Gemini 1.5 Flash | 92.3 | #620 |
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-major-object-recognition · How It Works · Data refreshed daily, snapshot 2026-11-05.