OpenVLM POPE: leaderboard
OpenCompass OpenVLM evaluation of POPE: polling-based object hallucination assessment for vision-language models, testing whether models falsely claim nonexistent objects are present.
Metric: Overall (%). Source: huggingface.co. Status: saturated. 216 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | InternVL3-78B | 90.5 |
| 2 | InternVL3-8B | 90.4 |
| 3 | InternVL3-38B | 89.2 |
| 4 | Gemini 1.5 Flash | 88.5 |
| 5 | Qwen 2 VL 7B | 88.4 |
| 6 | Gemini 1.5 Pro | 88.2 |
| 7 | Llama 3.2 11B Instruct | 88.1 |
| 8 | Qwen 2 VL 2B | 87.3 |
| 9 | Qwen 2 VL 72B | 87.2 |
| 10 | GPT-4.1 (2025-04-14) | 86.4 |
| 11 | Llama 3.2 90B Vision Instruct | 86.3 |
| 12 | GPT-4.1 Mini | 85.6 |
| 13 | Gemma 3 12B | 85.5 |
| 14 | Gemma 3 4B | 84.6 |
| 15 | InternVL2-8B | 84.2 |
Interactive version: theaggregate.ai/benchmark?slug=openvlm-pope · How It Works · Data refreshed daily, snapshot 2026-09-05.