multimodal benchmark
TVBench
TVBench is a temporal video understanding benchmark evaluating reasoning over actions, events, and temporal dynamics in videos.
Updated Aug 7, 2026
Published models3
Registry coverage3
MetricScore
EvidenceB
TVBench leaderboard
Sorted by the source-provided rank. Higher score is better according to the registry.
| 1 | BY | 80.5% | 100.0% | 3 | C | |
| 2 | AC | 78.2% | 50.0% | 3 | C | |
| 3 | BY | 77.2% | 0.0% | 3 | C |
Score distribution
Top published rows on the benchmark's original scale.
TVBench highlights
The top published results on this benchmark's own scale.
What is TVBench?
Definition and scoring fields from the benchmark registry.
TVBench is a temporal video understanding benchmark evaluating reasoning over actions, events, and temporal dynamics in videos.
Scores are shown in ratio. The current registry marks this benchmark as not independently verified with evidence level B.
- Family
- TVBench
- Modality
- multimodal
- Primary category
- multimodal
- Score direction
- higher
- LLMBoard eligible
- No
- Evaluation key
- tvbench|llm-stats-current
Source-native results are preserved. Eligibility for the overall LLMBoard score is a separate policy decision.
FAQ
Common questions about TVBench.
Which model scores highest on TVBench?
Seed 2.1 Pro is currently ranked first with 80.5%.
What does TVBench measure?
TVBench is a temporal video understanding benchmark evaluating reasoning over actions, events, and temporal dynamics in videos.
Is a higher score better?
Yes. Higher values rank better for this benchmark.
How many models are compared?
3 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.