multimodal benchmark
PerceptionTest
A novel multimodal video benchmark designed to evaluate perception and reasoning skills of pre-trained models across video, audio, and text modalities. Contains 11.6k real-world videos (average 23 seconds) filmed by participants worldwide, densely annotated with six types of labels. Focuses on skills (Memory, Abstraction, Physics, Semantics) and reasoning types (descriptive, explanatory, predictive, counterfactual). Shows significant performance gap between human baseline (91.4%) and state-of-the-art video QA models (46.2%).
Updated Aug 7, 2026
Published models2
Registry coverage2
MetricScore
EvidenceB
PerceptionTest leaderboard
Sorted by the source-provided rank. Higher score is better according to the registry.
| 1 | AC | 73.2% | 100.0% | 2 | C | |
| 2 | AC | 70.5% | 0.0% | 2 | C |
Score distribution
Top published rows on the benchmark's original scale.
PerceptionTest highlights
The top published results on this benchmark's own scale.
What is PerceptionTest?
Definition and scoring fields from the benchmark registry.
A novel multimodal video benchmark designed to evaluate perception and reasoning skills of pre-trained models across video, audio, and text modalities. Contains 11.6k real-world videos (average 23 seconds) filmed by participants worldwide, densely annotated with six types of labels. Focuses on skills (Memory, Abstraction, Physics, Semantics) and reasoning types (descriptive, explanatory, predictive, counterfactual). Shows significant performance gap between human baseline (91.4%) and state-of-the-art video QA models (46.2%).
Scores are shown in ratio. The current registry marks this benchmark as not independently verified with evidence level B.
- Family
- PerceptionTest
- Modality
- multimodal
- Primary category
- multimodal
- Score direction
- higher
- LLMBoard eligible
- No
- Evaluation key
- perceptiontest|llm-stats-current
Source-native results are preserved. Eligibility for the overall LLMBoard score is a separate policy decision.
FAQ
Common questions about PerceptionTest.
Which model scores highest on PerceptionTest?
Qwen2.5 VL 72B Instruct is currently ranked first with 73.2%.
What does PerceptionTest measure?
A novel multimodal video benchmark designed to evaluate perception and reasoning skills of pre-trained models across video, audio, and text modalities. Contains 11.6k real-world videos (average 23 seconds) filmed by participants worldwide, densely annotated with six types of labels. Focuses on skills (Memory, Abstraction, Physics, Semantics) and reasoning types (descriptive, explanatory, predictive, counterfactual). Shows significant performance gap between human baseline (91.4%) and state-of-the-art video QA models (46.2%).
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.