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: github.com. Source: huggingface.co. Saturation forecast: Estimated already saturated. 216 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | InternVL3-78B | 90.5 | #352 |
| 2 | InternVL3-8B | 90.4 | #620 |
| 3 | InternVL3-14B | 89.4 | #505 |
| 4 | InternVL3-38B | 89.2 | #405 |
| 5 | Gemini 1.5 Flash | 88.5 | #618 |
| 6 | Qwen 2 VL 7B | 88.4 | #750 |
| 7 | Gemini 1.5 Pro | 88.2 | #445 |
| 8 | Llama 3.2 11B Instruct | 88.1 | #1126 |
| 9 | Qwen 2 VL 2B | 87.3 | #1220 |
| 10 | Qwen 2 VL 72B | 87.2 | #460 |
| 11 | GPT-4.1 (2025-04-14) | 86.4 | #250 |
| 12 | Llama 3.2 90B Vision Instruct | 86.3 | #622 |
| 13 | Qwen 2.5 VL 7B | 85.9 | #643 |
| 14 | GPT-4.1 Mini | 85.6 | #355 |
| 15 | Gemma 3 12B | 85.5 | #671 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=openvlm-pope · How It Works · Data refreshed daily, snapshot 2026-10-20.