jina-v5-text-small added
The newest model in the Embedding Models Benchmarks is jina-v5-text-small.
Embedding Benchmarks
Transparent embedding models benchmarks built on AIMultiple's own retrieval evaluations of commercial and open-source embedding models across three enterprise RAG domains. See the methodology.
TOP MODEL
voyage-3.5
nDCG@3 0.943
Best value
pplx-0.6b
$0.0040 / 1M
Largest model
KaLM-Gemma3-12B
11.8B params
Coverage
29 models
3 retrieval domains · 87 evaluations
Embedding Models Benchmarks (nDCG@3)
Leaderboard
Every benchmarked embedding model, ranked by its three-domain nDCG@3 average.
Filter & Sort
# | Model | nDCG@3 | CUAD (legal) | TechQA (support) | MedRAG (health) |
|---|---|---|---|---|---|
| 1 |  voyage-3.5 Voyage AI | 0.943 | 0.910 | 0.965 | 0.954 |
| 2 | voyage-4-large Voyage AI | 0.942 | 0.873 | 0.966 | 0.986 |
| 3 |  gemini-2-preview Google | 0.932 | 0.896 | 0.930 | 0.969 |
| 4 |  nvidia-nemotron-8b NVIDIA | 0.925 | 0.860 | 0.952 | 0.963 |
| 5 | gemini-001 Google | 0.922 | 0.898 | 0.886 | 0.981 |
| 6 | voyage-4-lite Voyage AI | 0.921 | 0.859 | 0.948 | 0.956 |
| 7 | voyage-law-2 Voyage AI | 0.919 | 0.913 | 0.902 | 0.941 |
| 8 |  SFR-Embedding-2_R Salesforce | 0.905 | 0.842 | 0.911 | 0.962 |
| 9 |  jina-v5-text-small Jina AI | 0.895 | 0.836 | 0.897 | 0.952 |
| 10 |  harrier-oss-0.6b Microsoft | 0.891 | 0.872 | 0.841 | 0.961 |
Page 1 of 3 | |||||
Price$0.060
Params-
ReleasedMay 2025
CUAD (legal)
0.91
TechQA (support)
0.965
MedRAG (health)
0.954
Price$0.120
Params-
ReleasedJan 2026
CUAD (legal)
0.873
TechQA (support)
0.966
MedRAG (health)
0.986
Price$0.200
Params-
ReleasedMar 2026
CUAD (legal)
0.896
TechQA (support)
0.93
MedRAG (health)
0.969
Price$0.022
Params7.5B
ReleasedOct 2025
CUAD (legal)
0.86
TechQA (support)
0.952
MedRAG (health)
0.963
Price$0.150
Params-
ReleasedJul 2025
CUAD (legal)
0.898
TechQA (support)
0.886
MedRAG (health)
0.981
Price$0.020
Params-
ReleasedJan 2026
CUAD (legal)
0.859
TechQA (support)
0.948
MedRAG (health)
0.956
Price$0.120
Params-
ReleasedApr 2024
CUAD (legal)
0.913
TechQA (support)
0.902
MedRAG (health)
0.941
Price$0.041
Params7.1B
ReleasedJun 2024
CUAD (legal)
0.842
TechQA (support)
0.911
MedRAG (health)
0.962
Price$0.0066
Params1.6B
ReleasedApr 2026
CUAD (legal)
0.836
TechQA (support)
0.897
MedRAG (health)
0.952
Price$0.012
Params596M
ReleasedMar 2026
CUAD (legal)
0.872
TechQA (support)
0.841
MedRAG (health)
0.961
Page 1 of 3
Embedding models benchmarks by release date
nDCG@3 against each model's release date
Price vs Retrieval Accuracy
Effective cost against nDCG@3. Commercial models are priced at their API list rate, self-hosted models at the amortized GPU cost of running them on a single H100.
Methodology
How Embedding Models Benchmarks are Built
The embedding models benchmarks score each model by nDCG@3, the share of queries where the gold document lands in the top 3 results, and average its standing across the three retrieval domains it was evaluated on. Each benchmark below feeds that score.
Recent Updates
Recent Updates
Latest changes to the Embedding Models Benchmarks, model coverage and AIMultiple benchmark methodology.
Jina AI
jina-v5-text-small
New model added to the Embedding Models Benchmarks.
Microsoft
harrier-oss-0.6b
New model added to the Embedding Models Benchmarks.
Google
gemini-2-preview
New model added to the Embedding Models Benchmarks.
Voyage AI
voyage-4-large
New model added to the Embedding Models Benchmarks.
