multimodal benchmark
POPE
Polling-based Object Probing Evaluation (POPE) is a benchmark for evaluating object hallucination in Large Vision-Language Models (LVLMs). POPE addresses the problem where LVLMs generate objects inconsistent with target images by using a polling-based query method that asks yes/no questions about object presence in images, providing more stable and flexible evaluation of object hallucination.
Updated Aug 7, 2026
Published models2
Registry coverage2
MetricScore
EvidenceB
POPE leaderboard
Sorted by the source-provided rank. Higher score is better according to the registry.
| 1 | MI | 86.1% | 100.0% | 2 | C | |
| 2 | MI | 85.6% | 0.0% | 2 | C |
Score distribution
Top published rows on the benchmark's original scale.
POPE highlights
The top published results on this benchmark's own scale.
What is POPE?
Definition and scoring fields from the benchmark registry.
Polling-based Object Probing Evaluation (POPE) is a benchmark for evaluating object hallucination in Large Vision-Language Models (LVLMs). POPE addresses the problem where LVLMs generate objects inconsistent with target images by using a polling-based query method that asks yes/no questions about object presence in images, providing more stable and flexible evaluation of object hallucination.
Scores are shown in ratio. The current registry marks this benchmark as not independently verified with evidence level B.
- Family
- POPE
- Modality
- multimodal
- Primary category
- multimodal
- Score direction
- higher
- LLMBoard eligible
- No
- Evaluation key
- pope|llm-stats-current
Source-native results are preserved. Eligibility for the overall LLMBoard score is a separate policy decision.
FAQ
Common questions about POPE.
Which model scores highest on POPE?
Phi-3.5-vision-instruct is currently ranked first with 86.1%.
What does POPE measure?
Polling-based Object Probing Evaluation (POPE) is a benchmark for evaluating object hallucination in Large Vision-Language Models (LVLMs). POPE addresses the problem where LVLMs generate objects inconsistent with target images by using a polling-based query method that asks yes/no questions about object presence in images, providing more stable and flexible evaluation of object hallucination.
Is a higher score better?
Yes. Higher values rank better for this benchmark.
How many models are compared?
2 unique published model results are currently shown.
Does this benchmark affect the overall score?
This benchmark is preserved as source-native evidence but is not eligible for the current overall score.