multimodal benchmark
CharadesSTA
Charades-STA is a benchmark dataset for temporal activity localization via language queries, extending the Charades dataset with sentence temporal annotations. It contains 12,408 training and 3,720 testing segment-sentence pairs from videos with natural language descriptions and precise temporal boundaries for localizing activities based on language queries.
Updated Aug 7, 2026
Published models12
Registry coverage12
MetricScore
EvidenceB
CharadesSTA leaderboard
Sorted by the source-provided rank. Higher score is better according to the registry.
| 1 | AC | 64.8% | 100.0% | 12 | C | |
| 2 | AC | 63.5% | 90.9% | 12 | C | |
| 3 | AC | 63.5% | 81.8% | 12 | C | |
| 4 | AC | 62.8% | 72.7% | 12 | C | |
| 5 | AC | 62.7% | 63.6% | 12 | C | |
| 6 | AC | 61.2% | 54.5% | 12 | C | |
| 7 | AC | 59.9% | 45.5% | 12 | C | |
| 8 | AC | 59.0% | 36.4% | 12 | C | |
| 9 | AC | 56.0% | 27.3% | 12 | C | |
| 10 | AC | 55.5% | 18.2% | 12 | C | |
| 11 | AC | 54.2% | 9.1% | 12 | C | |
| 12 | AC | 43.6% | 0.0% | 12 | C |
Score distribution
Top published rows on the benchmark's original scale.
CharadesSTA highlights
The top published results on this benchmark's own scale.
What is CharadesSTA?
Definition and scoring fields from the benchmark registry.
Charades-STA is a benchmark dataset for temporal activity localization via language queries, extending the Charades dataset with sentence temporal annotations. It contains 12,408 training and 3,720 testing segment-sentence pairs from videos with natural language descriptions and precise temporal boundaries for localizing activities based on language queries.
Scores are shown in ratio. The current registry marks this benchmark as not independently verified with evidence level B.
- Family
- CharadesSTA
- Modality
- multimodal
- Primary category
- multimodal
- Score direction
- higher
- LLMBoard eligible
- No
- Evaluation key
- charadessta|llm-stats-current
Source-native results are preserved. Eligibility for the overall LLMBoard score is a separate policy decision.
FAQ
Common questions about CharadesSTA.
Which model scores highest on CharadesSTA?
Qwen3 VL 235B A22B Instruct is currently ranked first with 64.8%.
What does CharadesSTA measure?
Charades-STA is a benchmark dataset for temporal activity localization via language queries, extending the Charades dataset with sentence temporal annotations. It contains 12,408 training and 3,720 testing segment-sentence pairs from videos with natural language descriptions and precise temporal boundaries for localizing activities based on language queries.
Is a higher score better?
Yes. Higher values rank better for this benchmark.
How many models are compared?
12 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.