multimodal benchmark
TempCompass
TempCompass is a comprehensive benchmark for evaluating temporal perception capabilities of Video Large Language Models (Video LLMs). It constructs conflicting videos that share identical static content but differ in specific temporal aspects to prevent models from exploiting single-frame bias. The benchmark evaluates multiple temporal aspects including action, motion, speed, temporal order, and attribute changes across diverse task formats including multi-choice QA, yes/no QA, caption matching, and caption generation.
Updated Aug 7, 2026
Published models2
Registry coverage2
MetricScore
EvidenceB
TempCompass leaderboard
Sorted by the source-provided rank. Higher score is better according to the registry.
| 1 | AC | 74.8% | 100.0% | 2 | C | |
| 2 | AC | 71.7% | 0.0% | 2 | C |
Score distribution
Top published rows on the benchmark's original scale.
TempCompass highlights
The top published results on this benchmark's own scale.
What is TempCompass?
Definition and scoring fields from the benchmark registry.
TempCompass is a comprehensive benchmark for evaluating temporal perception capabilities of Video Large Language Models (Video LLMs). It constructs conflicting videos that share identical static content but differ in specific temporal aspects to prevent models from exploiting single-frame bias. The benchmark evaluates multiple temporal aspects including action, motion, speed, temporal order, and attribute changes across diverse task formats including multi-choice QA, yes/no QA, caption matching, and caption generation.
Scores are shown in ratio. The current registry marks this benchmark as not independently verified with evidence level B.
- Family
- TempCompass
- Modality
- multimodal
- Primary category
- multimodal
- Score direction
- higher
- LLMBoard eligible
- No
- Evaluation key
- tempcompass|llm-stats-current
Source-native results are preserved. Eligibility for the overall LLMBoard score is a separate policy decision.
FAQ
Common questions about TempCompass.
Which model scores highest on TempCompass?
Qwen2.5 VL 72B Instruct is currently ranked first with 74.8%.
What does TempCompass measure?
TempCompass is a comprehensive benchmark for evaluating temporal perception capabilities of Video Large Language Models (Video LLMs). It constructs conflicting videos that share identical static content but differ in specific temporal aspects to prevent models from exploiting single-frame bias. The benchmark evaluates multiple temporal aspects including action, motion, speed, temporal order, and attribute changes across diverse task formats including multi-choice QA, yes/no QA, caption matching, and caption generation.
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.