OpenVLM A-Bench - Nouns as Adjectives: leaderboard
A-Bench Nouns as Adjectives accuracy from the OpenCompass OpenVLM Leaderboard (VLMEvalKit): multiple-choice questions on AI-generated images, on prompts in which a noun modifies another object, so the model must not assume that noun's object is present. 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 | Qwen 2 VL 2B | 86 | #1191 |
| 2 | GPT-4.1 (2025-04-14) | 83.7 | #257 |
| 3 | InternVL3-78B | 81.4 | #361 |
| 4 | InternVL3-14B | 81.4 | #506 |
| 5 | Aquila-VL-2B | 81.4 | #1216 |
| 6 | MiniCPM-V-2.6 | 79.1 | #827 |
| 7 | Gemini 1.5 Pro | 79.1 | #454 |
| 8 | Gemma 3 4B | 79.1 | #1088 |
| 9 | Moonshot v1 8K | 79.1 | #622 |
| 10 | InternVL3-38B | 76.7 | #413 |
| 11 | InternVL2-8B | 76.7 | #813 |
| 12 | Claude 3.7 Sonnet | 76.7 | #259 |
| 13 | Llama 3.2 11B Instruct | 76.7 | #1108 |
| 14 | Gemini 1.5 Flash | 76.7 | #620 |
| 15 | Grok 2 (1212) | 76.7 | #502 |
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-nouns-as-adjectives · How It Works · Data refreshed daily, snapshot 2026-11-05.