Jina Embeddings v3
Built on XLM-RoBERTa with 570M parameters supporting 8,192 token context length across 89 languages. Features five task-specific LoRA adapters for retrieval, clustering, and classification with Matryoshka learning for flexible sizing.
Model Information
- Provider
- Jina AI
- License
- Open Source
- Price per 1M tokens
- $0.045
- Dimensions
- 1024
- Release Date
- 2024-09-18
- Model Name
- jina-embeddings-v3
- Total Evaluations
- 1918
Performance Record
Wins772 (40.3%)
Losses901 (47.0%)
Ties245 (12.8%)
Wins
Losses
Ties
Performance Overview
ELO ratings by dataset
Jina Embeddings v3's ELO performance varies across different benchmark datasets, showing its strengths in specific domains.
Jina Embeddings v3 - ELO by Dataset
Detailed Metrics
Dataset breakdown
Performance metrics across different benchmark datasets, including accuracy and latency percentiles.
ARCD
ELO 153047.1% WR113W-69L-58T
Accuracy Metrics
- nDCG@5
- 0.888
- nDCG@10
- 0.900
- Recall@5
- 0.920
- Recall@10
- 0.960
Latency Distribution
- Mean
- 236ms
- P50 (Median)
- 232ms
- P90
- 272ms
PG
ELO 150550.0% WR120W-115L-5T
Latency Distribution
- Mean
- 26ms
- P50 (Median)
- 26ms
- P90
- 30ms
MSMARCO
ELO 150444.6% WR107W-98L-35T
Accuracy Metrics
- nDCG@5
- 1.000
- nDCG@10
- 0.997
- Recall@5
- 0.123
- Recall@10
- 0.220
Latency Distribution
- Mean
- 92ms
- P50 (Median)
- 90ms
- P90
- 106ms
NorQuAD
ELO 149725.4% WR61W-68L-111T
Latency Distribution
- Mean
- 117ms
- P50 (Median)
- 115ms
- P90
- 135ms
FiQa
ELO 148442.5% WR102W-129L-9T
Accuracy Metrics
- nDCG@5
- 0.645
- nDCG@10
- 0.679
- Recall@5
- 0.633
- Recall@10
- 0.741
Latency Distribution
- Mean
- 89ms
- P50 (Median)
- 87ms
- P90
- 102ms
business reports
ELO 147238.8% WR93W-140L-7T
Latency Distribution
- Mean
- 25ms
- P50 (Median)
- 25ms
- P90
- 29ms
DBPedia
ELO 147237.4% WR89W-134L-15T
Accuracy Metrics
- nDCG@5
- 0.616
- nDCG@10
- 0.591
- Recall@5
- 0.225
- Recall@10
- 0.340
Latency Distribution
- Mean
- 39ms
- P50 (Median)
- 38ms
- P90
- 45ms
SciFact
ELO 146236.3% WR87W-148L-5T
Accuracy Metrics
- nDCG@5
- 0.622
- nDCG@10
- 0.661
- Recall@5
- 0.695
- Recall@10
- 0.800
Latency Distribution
- Mean
- 53ms
- P50 (Median)
- 52ms
- P90
- 61ms
Compare Models
See how it stacks up
Compare Jina Embeddings v3 with other top embeddings to understand the differences in performance, accuracy, and latency.