multimodal benchmark
Meld
MELD (Multimodal EmotionLines Dataset) is a multimodal multi-party dataset for emotion recognition in conversations. Contains approximately 13,000 utterances from 1,433 dialogues extracted from the TV series Friends. Each utterance is annotated with emotion (Anger, Disgust, Sadness, Joy, Neutral, Surprise, Fear) and sentiment labels across audio, visual, and textual modalities.
Updated Aug 7, 2026
Published models1
Registry coverage1
MetricScore
EvidenceB
Meld leaderboard
Sorted by the source-provided rank. Higher score is better according to the registry.
| 1 | AC | 57.0% | 100.0% | 1 | C |
Meld highlights
The top published results on this benchmark's own scale.
What is Meld?
Definition and scoring fields from the benchmark registry.
MELD (Multimodal EmotionLines Dataset) is a multimodal multi-party dataset for emotion recognition in conversations. Contains approximately 13,000 utterances from 1,433 dialogues extracted from the TV series Friends. Each utterance is annotated with emotion (Anger, Disgust, Sadness, Joy, Neutral, Surprise, Fear) and sentiment labels across audio, visual, and textual modalities.
Scores are shown in ratio. The current registry marks this benchmark as not independently verified with evidence level B.
- Family
- Meld
- Modality
- multimodal
- Primary category
- multimodal
- Score direction
- higher
- LLMBoard eligible
- No
- Evaluation key
- meld|llm-stats-current
Source-native results are preserved. Eligibility for the overall LLMBoard score is a separate policy decision.
FAQ
Common questions about Meld.
Which model scores highest on Meld?
Qwen2.5-Omni-7B is currently ranked first with 57.0%.
What does Meld measure?
MELD (Multimodal EmotionLines Dataset) is a multimodal multi-party dataset for emotion recognition in conversations. Contains approximately 13,000 utterances from 1,433 dialogues extracted from the TV series Friends. Each utterance is annotated with emotion (Anger, Disgust, Sadness, Joy, Neutral, Surprise, Fear) and sentiment labels across audio, visual, and textual modalities.
Is a higher score better?
Yes. Higher values rank better for this benchmark.
How many models are compared?
1 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.