OpenVLM CCBench - Scenery and Buildings: leaderboard
CCBench Scenery and Buildings accuracy from the OpenCompass OpenVLM Leaderboard (VLMEvalKit): multiple-choice questions on Chinese culture about scenery and buildings. The category holds 95 of the 510 questions and is scored with CircularEval (correct under every rotation of the options). Accuracy in percent, higher is better.
Metric: Accuracy (%). Source: huggingface.co. Saturation forecast: Estimated already saturated. 228 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-4.1 (2025-04-14) | 94.7 | #257 |
| 2 | Qwen 2 VL 72B | 88.4 | #450 |
| 3 | Qwen-VL-Max | 87.4 | #475 |
| 4 | InternVL3-78B | 84.2 | #361 |
| 5 | Qwen 2.5 VL 7B | 83.2 | #641 |
| 6 | GPT-4.1 Mini | 82.1 | #363 |
| 7 | Qwen 2 VL 7B | 77.9 | #743 |
| 8 | InternVL3-8B | 76.8 | #624 |
| 9 | InternVL3-38B | 76.8 | #413 |
| 10 | InternVL2-8B | 76.8 | #813 |
| 11 | Claude 3.5 Sonnet (20241022) | 74.7 | #302 |
| 12 | Moonshot v1 8K | 72.6 | #622 |
| 13 | InternVL3-14B | 70.5 | #506 |
| 14 | Llama 3.2 90B Vision Instruct | 70.5 | #621 |
| 15 | Grok 2 (1212) | 69.5 | #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-ccbench-scenery-and-buildings · How It Works · Data refreshed daily, snapshot 2026-11-05.