Jina AI: Jina Embeddings v4

Jina Embeddings V4 is a 3.8 billion parameter multimodal embedding model that provides unified text and image representation capabilities. Built on the Qwen2.5-VL-3B-Instruct backbone, the model feat…

jina/jina-embeddings-v4

Jina Embeddings V4 is a 3.8 billion parameter multimodal embedding model that provides unified text and image representation capabilities. Built on the Qwen2.5-VL-3B-Instruct backbone, the model features an architecture that supports both single-vector and multi-vector embeddings in the late interaction style, addressing limitations found in traditional CLIP-style dual-encoder models. The model incorporates three specialized task-specific LoRA adapters (60M parameters each) that optimize performance across different retrieval scenarios including asymmetric query-document retrieval, semantic text similarity, and code search without modifying the frozen backbone weights. The model demonstrates strong performance in processing visually rich content such as tables, charts, diagrams, screenshots, and mixed-media formats through a unified processing pathway that reduces the modality gap present in conventional architectures. Supporting multilingual capabilities, the model can handle input texts up to 32,768 tokens with images resized to 20 megapixels, making it suitable for various document retrieval and cross-modal search applications across different languages and domains.

Specifications & capabilities

Input
text, image
Output
embeddings
Context window · tokens
32768
Embedding dimensions
2048
Published model ID
jina-embeddings-v4
Capabilities
embeddings, multimodal, visionInput

Evaluations

1–15 / 15

MTEB

89.889%

source id
mteb
leaderboard name
MTEB Bitext Mining
dimension
embedding
metric
Average task score
rank
7
sample size
5
aggregation scope
splits
  • default
  • test
subsets
  • ces-slk
  • default
  • eng-slk
task_count
3
task_names
  • OpusSlovakEnglishBitextMining
  • WebFAQBitextMiningQAs
  • WebFAQBitextMiningQuestions
sample_size
5
subset_count
3
subset_sample
  • ces-slk
  • default
  • eng-slk
aggregation_method
mean-of-task-means
model_revision_count
1
model_revision_sample
  • 4a58ca57710c49f51896e4bc820e202fbf64904b
source model id
jinaai/jina-embeddings-v4
Original result ↗

MTEB

85.246%

source id
mteb
leaderboard name
MTEB Reranking
dimension
embedding
metric
Average task score
rank
22
sample size
3
aggregation scope
splits
  • test
subsets
  • default
task_count
3
task_names
  • SkQuadReranking
  • SlovakPharmacyDrMaxReranking
  • SlovakPharmacyMojaLekarenReranking
sample_size
3
subset_count
1
subset_sample
  • default
aggregation_method
mean-of-task-means
model_revision_count
1
model_revision_sample
  • 4a58ca57710c49f51896e4bc820e202fbf64904b
source model id
jinaai/jina-embeddings-v4
Original result ↗

MTEB

85.654%

source id
mteb
leaderboard name
MTEB Semantic Similarity
dimension
embedding
metric
Average task score
rank
40
sample size
2
aggregation scope
splits
  • test
subsets
  • default
task_count
2
task_names
  • SlovakSTS
  • SlovakSumSTS
sample_size
2
subset_count
1
subset_sample
  • default
aggregation_method
mean-of-task-means
model_revision_count
1
model_revision_sample
  • 4a58ca57710c49f51896e4bc820e202fbf64904b
source model id
jinaai/jina-embeddings-v4
Original result ↗

MTEB

59.937%

source id
mteb
leaderboard name
MTEB Summarization
dimension
embedding
metric
Average task score
rank
53
sample size
1
aggregation scope
splits
  • test
subsets
  • default
task_count
1
task_names
  • LegalSummarization
sample_size
1
subset_count
1
subset_sample
  • default
aggregation_method
mean-of-task-means
model_revision_count
1
model_revision_sample
  • 4a58ca57710c49f51896e4bc820e202fbf64904b
source model id
jinaai/jina-embeddings-v4
Original result ↗

MTEB

67.003%

source id
mteb
leaderboard name
MTEB Overall
dimension
embedding
metric
Average task score
rank
150
sample size
361
aggregation scope
splits
  • dev
  • test
  • test_1024
  • test_16384
  • test_2048
  • test_256
  • test_32768
  • test_4096
  • test_512
  • test_8192
  • train
  • validation
subsets
  • afr
  • amh
  • ar
  • ar-ar
  • ara-ara
  • ara-deu
  • ara-eng
  • ara-hin
  • ara-spa
  • ara-vie
  • ara-zho
  • arb
  • arq
  • ary
  • bg
  • bn
  • cs
  • da
  • de
  • de-en
  • de-fr
  • de-pl
  • default
  • deu-ara
  • deu-deu
  • deu-eng
  • deu-hin
  • deu-spa
  • deu-vie
  • deu-zho
  • en
  • en-ar
  • en-as
  • en-bn
  • en-de
  • en-en
  • en-gu
  • en-hi
  • en-kn
  • en-ml
  • en-mr
  • en-or
  • en-pa
  • en-ta
  • en-te
  • en-tr
  • en-ur
  • eng
  • eng-ara
  • eng-deu
  • eng-eng
  • eng-hin
  • eng-spa
  • eng-vie
  • eng-zho
  • english
  • es
  • es-en
  • es-es
  • es-it
  • fa
  • fi
  • fr
  • fr-en
  • fr-pl
  • french
  • german
  • go
  • hau
  • hi
  • hin
  • hin-ara
  • hin-deu
  • hin-eng
  • hin-hin
  • hin-spa
  • hin-vie
  • hin-zho
  • hu
  • id
  • ind
  • it
  • it-en
  • java
  • javascript
  • jp
  • kin
  • ko
  • ko-ko
  • mar
  • my
  • nl
  • nl-en
  • no
  • php
  • pl
  • pl-en
  • pt
  • python
  • ro
  • ru
  • ruby
  • spa-ara
  • spa-deu
  • spa-eng
  • spa-hin
  • spa-spa
  • spa-vie
  • spa-zho
  • spanish
  • sr
  • sv
  • tel
  • th
  • tr
  • ur
  • vi
  • vie-ara
  • vie-deu
  • vie-eng
  • vie-hin
  • vie-spa
  • vie-vie
  • vie-zho
  • zh
  • zh-en
  • zho-ara
  • zho-deu
  • zho-eng
  • zho-hin
  • zho-spa
  • zho-vie
  • zho-zho
task_count
109
task_names
  • AlloprofRetrieval
  • ArguAna
  • BIOSSES
  • COIRCodeSearchNetRetrieval
  • CQADupstackAndroidRetrieval
  • CQADupstackEnglishRetrieval
  • CQADupstackGamingRetrieval
  • CQADupstackGisRetrieval
  • CQADupstackMathematicaRetrieval
  • CQADupstackPhysicsRetrieval
  • CQADupstackProgrammersRetrieval
  • CQADupstackStatsRetrieval
  • CQADupstackTexRetrieval
  • CQADupstackUnixRetrieval
  • CQADupstackWebmastersRetrieval
  • CQADupstackWordpressRetrieval
  • ChemNQRetrieval
  • ClimateFEVER
  • ClimateFEVERHardNegatives
  • CodeSearchNetCCRetrieval
  • CodeTransOceanContest
  • CodeTransOceanDL
  • CosQA
  • CovidRetrieval
  • DBPedia
  • FEVERHardNegatives
  • FaroeseSTS
  • FiQA2018
  • FinParaSTS
  • HagridRetrieval
  • HotpotQAHardNegatives
  • IndicCrosslingualSTS
  • JSICK
  • JaGovFaqsRetrieval
  • JinaVDRAirbnbSyntheticRetrieval
  • JinaVDRArabicChartQARetrieval
  • JinaVDRArabicInfographicsVQARetrieval
  • JinaVDRAutomobileCatelogRetrieval
  • JinaVDRBeveragesCatalogueRetrieval
  • JinaVDRCharXivOCRRetrieval
  • JinaVDRChartQARetrieval
  • JinaVDRDonutVQAISynHMPRetrieval
  • JinaVDREuropeanaDeNewsRetrieval
  • JinaVDREuropeanaEsNewsRetrieval
  • JinaVDREuropeanaFrNewsRetrieval
  • JinaVDREuropeanaItScansRetrieval
  • JinaVDREuropeanaNlLegalRetrieval
  • JinaVDRGitHubReadmeRetrieval
  • JinaVDRHindiGovVQARetrieval
  • JinaVDRHungarianDocQARetrieval
  • JinaVDRJDocQARetrieval
  • JinaVDRJina2024YearlyBookRetrieval
  • JinaVDRMMTabRetrieval
  • JinaVDRMPMQARetrieval
  • JinaVDRMedicalPrescriptionsRetrieval
  • JinaVDROWIDChartsRetrieval
  • JinaVDROpenAINewsRetrieval
  • JinaVDRPlotQARetrieval
  • JinaVDRRamensBenchmarkRetrieval
  • JinaVDRShanghaiMasterPlanRetrieval
  • JinaVDRShiftProjectRetrieval
  • JinaVDRStanfordSlideRetrieval
  • JinaVDRStudentEnrollmentSyntheticRetrieval
  • JinaVDRTQARetrieval
  • JinaVDRTabFQuadRetrieval
  • JinaVDRTableVQARetrieval
  • JinaVDRTweetStockSyntheticsRetrieval
  • JinaVDRWikimediaCommonsDocumentsRetrieval
  • JinaVDRWikimediaCommonsMapsRetrieval
  • LEMBNarrativeQARetrieval
  • LEMBNeedleRetrieval
  • LEMBPasskeyRetrieval
  • LEMBQMSumRetrieval
  • LEMBSummScreenFDRetrieval
  • LEMBWikimQARetrieval
  • LeCaRDv2
  • LegalBenchCorporateLobbying
  • MLQARetrieval
  • MLQuestions
  • MedicalQARetrieval
  • NFCorpus
  • QuoraRetrieval
  • RuBQRetrieval
  • SCIDOCS
  • STS13
  • STS14
  • STS15
  • STS17
  • STS22.v2
  • STSBenchmark
  • STSES
  • SciFact
  • SemRel24STS
  • SpartQA
  • StatcanDialogueDatasetRetrieval
  • SyntheticText2SQL
  • TRECCOVID
  • TempReasonL1
  • Touche2020
  • Touche2020Retrieval.v3
  • TwitterHjerneRetrieval
  • Vidore2BioMedicalLecturesRetrieval
  • Vidore2ESGReportsHLRetrieval
  • Vidore2ESGReportsRetrieval
  • Vidore2EconomicsReportsRetrieval
  • VidoreShiftProjectRetrieval
  • VidoreTabfquadRetrieval
  • WikipediaRetrievalMultilingual
  • WinoGrande
sample_size
361
subset_count
133
subset_sample
  • afr
  • amh
  • ar
  • ar-ar
  • ara-ara
  • ara-deu
  • ara-eng
  • ara-hin
  • ara-spa
  • ara-vie
  • ara-zho
  • arb
  • arq
  • ary
  • bg
  • bn
  • cs
  • da
  • de
  • de-en
  • de-fr
  • de-pl
  • default
  • deu-ara
  • deu-deu
  • deu-eng
  • deu-hin
  • deu-spa
  • deu-vie
  • deu-zho
aggregation_method
mean-of-task-means
model_revision_count
1
model_revision_sample
  • 26239889730c735ed7e9a4db9180c8935faf4ba0
source model id
jinaai/jina-embeddings-v4
Original result ↗

MTEB

70.123%

source id
mteb
leaderboard name
MTEB Retrieval
dimension
embedding
metric
Average task score
rank
157
sample size
275
aggregation scope
splits
  • dev
  • test
  • test_1024
  • test_16384
  • test_2048
  • test_256
  • test_32768
  • test_4096
  • test_512
  • test_8192
  • train
  • validation
subsets
  • ar
  • ara-ara
  • ara-deu
  • ara-eng
  • ara-hin
  • ara-spa
  • ara-vie
  • ara-zho
  • bg
  • bn
  • cs
  • da
  • de
  • default
  • deu-ara
  • deu-deu
  • deu-eng
  • deu-hin
  • deu-spa
  • deu-vie
  • deu-zho
  • en
  • eng-ara
  • eng-deu
  • eng-eng
  • eng-hin
  • eng-spa
  • eng-vie
  • eng-zho
  • english
  • es
  • fa
  • fi
  • fr
  • french
  • german
  • go
  • hi
  • hin-ara
  • hin-deu
  • hin-eng
  • hin-hin
  • hin-spa
  • hin-vie
  • hin-zho
  • hu
  • id
  • it
  • java
  • javascript
  • jp
  • ko
  • my
  • nl
  • no
  • php
  • pt
  • python
  • ro
  • ru
  • ruby
  • spa-ara
  • spa-deu
  • spa-eng
  • spa-hin
  • spa-spa
  • spa-vie
  • spa-zho
  • spanish
  • sr
  • sv
  • th
  • ur
  • vi
  • vie-ara
  • vie-deu
  • vie-eng
  • vie-hin
  • vie-spa
  • vie-vie
  • vie-zho
  • zh
  • zho-ara
  • zho-deu
  • zho-eng
  • zho-hin
  • zho-spa
  • zho-vie
  • zho-zho
task_count
74
task_names
  • AlloprofRetrieval
  • COIRCodeSearchNetRetrieval
  • CQADupstackAndroidRetrieval
  • CQADupstackEnglishRetrieval
  • CQADupstackGamingRetrieval
  • CQADupstackGisRetrieval
  • CQADupstackMathematicaRetrieval
  • CQADupstackPhysicsRetrieval
  • CQADupstackProgrammersRetrieval
  • CQADupstackStatsRetrieval
  • CQADupstackTexRetrieval
  • CQADupstackUnixRetrieval
  • CQADupstackWebmastersRetrieval
  • CQADupstackWordpressRetrieval
  • ChemNQRetrieval
  • CodeSearchNetCCRetrieval
  • CovidRetrieval
  • HagridRetrieval
  • JaGovFaqsRetrieval
  • JinaVDRAirbnbSyntheticRetrieval
  • JinaVDRArabicChartQARetrieval
  • JinaVDRArabicInfographicsVQARetrieval
  • JinaVDRAutomobileCatelogRetrieval
  • JinaVDRBeveragesCatalogueRetrieval
  • JinaVDRCharXivOCRRetrieval
  • JinaVDRChartQARetrieval
  • JinaVDRDonutVQAISynHMPRetrieval
  • JinaVDREuropeanaDeNewsRetrieval
  • JinaVDREuropeanaEsNewsRetrieval
  • JinaVDREuropeanaFrNewsRetrieval
  • JinaVDREuropeanaItScansRetrieval
  • JinaVDREuropeanaNlLegalRetrieval
  • JinaVDRGitHubReadmeRetrieval
  • JinaVDRHindiGovVQARetrieval
  • JinaVDRHungarianDocQARetrieval
  • JinaVDRJDocQARetrieval
  • JinaVDRJina2024YearlyBookRetrieval
  • JinaVDRMMTabRetrieval
  • JinaVDRMPMQARetrieval
  • JinaVDRMedicalPrescriptionsRetrieval
  • JinaVDROWIDChartsRetrieval
  • JinaVDROpenAINewsRetrieval
  • JinaVDRPlotQARetrieval
  • JinaVDRRamensBenchmarkRetrieval
  • JinaVDRShanghaiMasterPlanRetrieval
  • JinaVDRShiftProjectRetrieval
  • JinaVDRStanfordSlideRetrieval
  • JinaVDRStudentEnrollmentSyntheticRetrieval
  • JinaVDRTQARetrieval
  • JinaVDRTabFQuadRetrieval
  • JinaVDRTableVQARetrieval
  • JinaVDRTweetStockSyntheticsRetrieval
  • JinaVDRWikimediaCommonsDocumentsRetrieval
  • JinaVDRWikimediaCommonsMapsRetrieval
  • LEMBNarrativeQARetrieval
  • LEMBNeedleRetrieval
  • LEMBPasskeyRetrieval
  • LEMBQMSumRetrieval
  • LEMBSummScreenFDRetrieval
  • LEMBWikimQARetrieval
  • MLQARetrieval
  • MedicalQARetrieval
  • QuoraRetrieval
  • RuBQRetrieval
  • StatcanDialogueDatasetRetrieval
  • Touche2020Retrieval.v3
  • TwitterHjerneRetrieval
  • Vidore2BioMedicalLecturesRetrieval
  • Vidore2ESGReportsHLRetrieval
  • Vidore2ESGReportsRetrieval
  • Vidore2EconomicsReportsRetrieval
  • VidoreShiftProjectRetrieval
  • VidoreTabfquadRetrieval
  • WikipediaRetrievalMultilingual
sample_size
275
subset_count
89
subset_sample
  • ar
  • ara-ara
  • ara-deu
  • ara-eng
  • ara-hin
  • ara-spa
  • ara-vie
  • ara-zho
  • bg
  • bn
  • cs
  • da
  • de
  • default
  • deu-ara
  • deu-deu
  • deu-eng
  • deu-hin
  • deu-spa
  • deu-vie
  • deu-zho
  • en
  • eng-ara
  • eng-deu
  • eng-eng
  • eng-hin
  • eng-spa
  • eng-vie
  • eng-zho
  • english
aggregation_method
mean-of-task-means
model_revision_count
1
model_revision_sample
  • 26239889730c735ed7e9a4db9180c8935faf4ba0
source model id
jinaai/jina-embeddings-v4
Original result ↗

MTEB

64.039%

source id
mteb
leaderboard name
MTEB Other
dimension
embedding
metric
Average task score
rank
157
sample size
40
aggregation scope
splits
  • all
  • dentistry_and_oral_health
  • dermatology
  • gastroenterology
  • genetics
  • neuroscience_and_neurology
  • orthopedic_surgery
  • otorhinolaryngology
  • plastic_surgery
  • psychiatry_and_psychology
  • pulmonology
  • test
subsets
  • default
  • en
  • es
  • fr
  • slk-eng
task_count
8
task_names
  • AILACasedocs
  • AILAStatutes
  • CUREv1
  • DemagogSKNLI
  • LegalQuAD
  • SlovakNLI
  • SlovakRTE
  • Tatoeba
sample_size
40
subset_count
5
subset_sample
  • default
  • en
  • es
  • fr
  • slk-eng
aggregation_method
mean-of-task-means
model_revision_count
1
model_revision_sample
  • 4a58ca57710c49f51896e4bc820e202fbf64904b
source model id
jinaai/jina-embeddings-v4
Original result ↗

MTEB

63.91%

source id
mteb
leaderboard name
MTEB Overall
dimension
embedding
metric
Average task score
rank
222
sample size
189
aggregation scope
splits
  • all
  • default
  • dentistry_and_oral_health
  • dermatology
  • dev
  • gastroenterology
  • genetics
  • neuroscience_and_neurology
  • orthopedic_surgery
  • otorhinolaryngology
  • plastic_surgery
  • psychiatry_and_psychology
  • pulmonology
  • test
  • train
  • validation
subsets
  • ar
  • bn
  • ces-slk
  • de
  • default
  • en
  • eng-slk
  • eng_Latn-slk_Latn
  • english
  • es
  • fa
  • fi
  • fr
  • french
  • german
  • hi
  • id
  • italian
  • ja
  • ko
  • portuguese
  • ru
  • slk
  • slk-eng
  • slk_Latn
  • slk_Latn-eng_Latn
  • slk_Latn-slk_Latn
  • spanish
  • sw
  • te
  • th
  • yo
  • zh
task_count
59
task_names
  • AILACasedocs
  • AILAStatutes
  • BelebeleRetrieval
  • CUREv1
  • ChatDoctorRetrieval
  • DS1000Retrieval
  • DemagogSKNLI
  • FinQARetrieval
  • FinanceBenchRetrieval
  • FreshStackRetrieval
  • HC3FinanceRetrieval
  • HumanEvalRetrieval
  • LegalQuAD
  • LegalSummarization
  • MBPPRetrieval
  • MIRACLRetrievalHardNegatives
  • MultiEupSlovakGenderClassification
  • MultiEupSlovakPartyClassification
  • MultilingualSentimentClassification
  • OpusSlovakEnglishBitextMining
  • PravdaSKTagClustering
  • PravdaSKURLClustering
  • SIB200Classification
  • SIB200ClusteringS2S
  • SKQuadRetrieval
  • SMESumCategoryClustering
  • SMESumRetrieval
  • SkQuadReranking
  • SlovakHateSpeechClassification.v2
  • SlovakMovieReviewSentimentClassification.v2
  • SlovakNLI
  • SlovakParlaSentClassification
  • SlovakPharmacyDrMaxReranking
  • SlovakPharmacyMojaLekarenReranking
  • SlovakRTE
  • SlovakSTS
  • SlovakSumRetrieval
  • SlovakSumSTS
  • SlovakSumURLClustering
  • Tatoeba
  • Vidore3ComputerScienceRetrieval
  • Vidore3ComputerScienceRetrieval.v2
  • Vidore3EnergyRetrieval
  • Vidore3EnergyRetrieval.v2
  • Vidore3FinanceEnRetrieval
  • Vidore3FinanceEnRetrieval.v2
  • Vidore3FinanceFrRetrieval
  • Vidore3FinanceFrRetrieval.v2
  • Vidore3HrRetrieval
  • Vidore3HrRetrieval.v2
  • Vidore3IndustrialRetrieval
  • Vidore3IndustrialRetrieval.v2
  • Vidore3PharmaceuticalsRetrieval
  • Vidore3PharmaceuticalsRetrieval.v2
  • Vidore3PhysicsRetrieval
  • Vidore3PhysicsRetrieval.v2
  • WebFAQBitextMiningQAs
  • WebFAQBitextMiningQuestions
  • WikiSQLRetrieval
sample_size
189
subset_count
33
subset_sample
  • ar
  • bn
  • ces-slk
  • de
  • default
  • en
  • eng-slk
  • eng_Latn-slk_Latn
  • english
  • es
  • fa
  • fi
  • fr
  • french
  • german
  • hi
  • id
  • italian
  • ja
  • ko
  • portuguese
  • ru
  • slk
  • slk-eng
  • slk_Latn
  • slk_Latn-eng_Latn
  • slk_Latn-slk_Latn
  • spanish
  • sw
  • te
aggregation_method
mean-of-task-means
model_revision_count
1
model_revision_sample
  • 4a58ca57710c49f51896e4bc820e202fbf64904b
source model id
jinaai/jina-embeddings-v4
Original result ↗

MTEB

62.787%

source id
mteb
leaderboard name
MTEB Classification
dimension
embedding
metric
Average task score
rank
274
sample size
9
aggregation scope
splits
  • test
  • train
  • validation
subsets
  • default
  • slk
  • slk_Latn
task_count
7
task_names
  • MultiEupSlovakGenderClassification
  • MultiEupSlovakPartyClassification
  • MultilingualSentimentClassification
  • SIB200Classification
  • SlovakHateSpeechClassification.v2
  • SlovakMovieReviewSentimentClassification.v2
  • SlovakParlaSentClassification
sample_size
9
subset_count
3
subset_sample
  • default
  • slk
  • slk_Latn
aggregation_method
mean-of-task-means
model_revision_count
1
model_revision_sample
  • 4a58ca57710c49f51896e4bc820e202fbf64904b
source model id
jinaai/jina-embeddings-v4
Original result ↗

MTEB

44.646%

source id
mteb
leaderboard name
MTEB Clustering
dimension
embedding
metric
Average task score
rank
298
sample size
5
aggregation scope
splits
  • test
subsets
  • default
  • slk_Latn
task_count
5
task_names
  • PravdaSKTagClustering
  • PravdaSKURLClustering
  • SIB200ClusteringS2S
  • SMESumCategoryClustering
  • SlovakSumURLClustering
sample_size
5
subset_count
2
subset_sample
  • default
  • slk_Latn
aggregation_method
mean-of-task-means
model_revision_count
1
model_revision_sample
  • 4a58ca57710c49f51896e4bc820e202fbf64904b
source model id
jinaai/jina-embeddings-v4
Original result ↗

MTEB

61.299%

source id
mteb
leaderboard name
MTEB Retrieval
dimension
embedding
metric
Average task score
rank
307
sample size
124
aggregation scope
splits
  • dev
  • test
subsets
  • ar
  • bn
  • de
  • default
  • en
  • eng_Latn-slk_Latn
  • english
  • es
  • fa
  • fi
  • fr
  • french
  • german
  • hi
  • id
  • italian
  • ja
  • ko
  • portuguese
  • ru
  • slk_Latn-eng_Latn
  • slk_Latn-slk_Latn
  • spanish
  • sw
  • te
  • th
  • yo
  • zh
task_count
30
task_names
  • BelebeleRetrieval
  • ChatDoctorRetrieval
  • DS1000Retrieval
  • FinQARetrieval
  • FinanceBenchRetrieval
  • FreshStackRetrieval
  • HC3FinanceRetrieval
  • HumanEvalRetrieval
  • MBPPRetrieval
  • MIRACLRetrievalHardNegatives
  • SKQuadRetrieval
  • SMESumRetrieval
  • SlovakSumRetrieval
  • Vidore3ComputerScienceRetrieval
  • Vidore3ComputerScienceRetrieval.v2
  • Vidore3EnergyRetrieval
  • Vidore3EnergyRetrieval.v2
  • Vidore3FinanceEnRetrieval
  • Vidore3FinanceEnRetrieval.v2
  • Vidore3FinanceFrRetrieval
  • Vidore3FinanceFrRetrieval.v2
  • Vidore3HrRetrieval
  • Vidore3HrRetrieval.v2
  • Vidore3IndustrialRetrieval
  • Vidore3IndustrialRetrieval.v2
  • Vidore3PharmaceuticalsRetrieval
  • Vidore3PharmaceuticalsRetrieval.v2
  • Vidore3PhysicsRetrieval
  • Vidore3PhysicsRetrieval.v2
  • WikiSQLRetrieval
sample_size
124
subset_count
28
subset_sample
  • ar
  • bn
  • de
  • default
  • en
  • eng_Latn-slk_Latn
  • english
  • es
  • fa
  • fi
  • fr
  • french
  • german
  • hi
  • id
  • italian
  • ja
  • ko
  • portuguese
  • ru
  • slk_Latn-eng_Latn
  • slk_Latn-slk_Latn
  • spanish
  • sw
  • te
  • th
  • yo
  • zh
aggregation_method
mean-of-task-means
model_revision_count
1
model_revision_sample
  • 4a58ca57710c49f51896e4bc820e202fbf64904b
source model id
jinaai/jina-embeddings-v4
Original result ↗

MTEB

56.316%

source id
mteb
leaderboard name
MTEB Other
dimension
embedding
metric
Average task score
rank
329
sample size
25
aggregation scope
splits
  • dev
  • test
subsets
  • default
task_count
24
task_names
  • ArguAna
  • BIOSSES
  • ClimateFEVER
  • ClimateFEVERHardNegatives
  • CodeTransOceanContest
  • CodeTransOceanDL
  • CosQA
  • DBPedia
  • FEVERHardNegatives
  • FiQA2018
  • HotpotQAHardNegatives
  • JSICK
  • LeCaRDv2
  • LegalBenchCorporateLobbying
  • MLQuestions
  • NFCorpus
  • SCIDOCS
  • SciFact
  • SpartQA
  • SyntheticText2SQL
  • TRECCOVID
  • TempReasonL1
  • Touche2020
  • WinoGrande
sample_size
25
subset_count
1
subset_sample
  • default
aggregation_method
mean-of-task-means
model_revision_count
1
model_revision_sample
  • 26239889730c735ed7e9a4db9180c8935faf4ba0
source model id
jinaai/jina-embeddings-v4
Original result ↗

MTEB

69.336%

source id
mteb
leaderboard name
MTEB Semantic Similarity
dimension
embedding
metric
Average task score
rank
343
sample size
61
aggregation scope
splits
  • test
  • train
  • validation
subsets
  • afr
  • amh
  • ar
  • ar-ar
  • arb
  • arq
  • ary
  • de
  • de-en
  • de-fr
  • de-pl
  • default
  • en
  • en-ar
  • en-as
  • en-bn
  • en-de
  • en-en
  • en-gu
  • en-hi
  • en-kn
  • en-ml
  • en-mr
  • en-or
  • en-pa
  • en-ta
  • en-te
  • en-tr
  • en-ur
  • eng
  • es
  • es-en
  • es-es
  • es-it
  • fr
  • fr-en
  • fr-pl
  • hau
  • hin
  • ind
  • it
  • it-en
  • kin
  • ko-ko
  • mar
  • nl-en
  • pl
  • pl-en
  • ru
  • tel
  • tr
  • zh
  • zh-en
task_count
11
task_names
  • FaroeseSTS
  • FinParaSTS
  • IndicCrosslingualSTS
  • STS13
  • STS14
  • STS15
  • STS17
  • STS22.v2
  • STSBenchmark
  • STSES
  • SemRel24STS
sample_size
61
subset_count
53
subset_sample
  • afr
  • amh
  • ar
  • ar-ar
  • arb
  • arq
  • ary
  • de
  • de-en
  • de-fr
  • de-pl
  • default
  • en
  • en-ar
  • en-as
  • en-bn
  • en-de
  • en-en
  • en-gu
  • en-hi
  • en-kn
  • en-ml
  • en-mr
  • en-or
  • en-pa
  • en-ta
  • en-te
  • en-tr
  • en-ur
  • eng
aggregation_method
mean-of-task-means
model_revision_count
1
model_revision_sample
  • 26239889730c735ed7e9a4db9180c8935faf4ba0
source model id
jinaai/jina-embeddings-v4
Original result ↗

MTEB

56.637%

source id
mteb
leaderboard name
MTEB Retrieval
dimension
embedding
metric
Average task score
rank
364
sample size
91
aggregation scope
splits
  • test
subsets
  • english
  • french
  • german
  • italian
  • portuguese
  • spanish
task_count
16
task_names
  • Vidore3ComputerScienceRetrieval
  • Vidore3ComputerScienceRetrieval.v2
  • Vidore3EnergyRetrieval
  • Vidore3EnergyRetrieval.v2
  • Vidore3FinanceEnRetrieval
  • Vidore3FinanceEnRetrieval.v2
  • Vidore3FinanceFrRetrieval
  • Vidore3FinanceFrRetrieval.v2
  • Vidore3HrRetrieval
  • Vidore3HrRetrieval.v2
  • Vidore3IndustrialRetrieval
  • Vidore3IndustrialRetrieval.v2
  • Vidore3PharmaceuticalsRetrieval
  • Vidore3PharmaceuticalsRetrieval.v2
  • Vidore3PhysicsRetrieval
  • Vidore3PhysicsRetrieval.v2
experiments
model_type
  • late-interaction
vector_type
multi_vector
sample_size
91
subset_count
6
subset_sample
  • english
  • french
  • german
  • italian
  • portuguese
  • spanish
aggregation_method
mean-of-task-means
model_revision_count
1
model_revision_sample
  • 4a58ca57710c49f51896e4bc820e202fbf64904b
source model id
jinaai/jina-embeddings-v4
Original result ↗

MTEB

56.637%

source id
mteb
leaderboard name
MTEB Overall
dimension
embedding
metric
Average task score
rank
406
sample size
91
aggregation scope
splits
  • test
subsets
  • english
  • french
  • german
  • italian
  • portuguese
  • spanish
task_count
16
task_names
  • Vidore3ComputerScienceRetrieval
  • Vidore3ComputerScienceRetrieval.v2
  • Vidore3EnergyRetrieval
  • Vidore3EnergyRetrieval.v2
  • Vidore3FinanceEnRetrieval
  • Vidore3FinanceEnRetrieval.v2
  • Vidore3FinanceFrRetrieval
  • Vidore3FinanceFrRetrieval.v2
  • Vidore3HrRetrieval
  • Vidore3HrRetrieval.v2
  • Vidore3IndustrialRetrieval
  • Vidore3IndustrialRetrieval.v2
  • Vidore3PharmaceuticalsRetrieval
  • Vidore3PharmaceuticalsRetrieval.v2
  • Vidore3PhysicsRetrieval
  • Vidore3PhysicsRetrieval.v2
experiments
model_type
  • late-interaction
vector_type
multi_vector
sample_size
91
subset_count
6
subset_sample
  • english
  • french
  • german
  • italian
  • portuguese
  • spanish
aggregation_method
mean-of-task-means
model_revision_count
1
model_revision_sample
  • 4a58ca57710c49f51896e4bc820e202fbf64904b
source model id
jinaai/jina-embeddings-v4
Original result ↗

Published prices

1–1 / 1

Provider specifications & access

1–1 / 1

jina ↗

source model id
jina-ai/jina-embeddings-v4
hosting
jina-hosted
platforms
  • Jina AI
context length
32768
architecture
input_modalities
  • text
  • image
output_modalities
  • embeddings
capabilities
  • embeddings

Source specification observations

1–2 / 2

jina ↗

field
context_length
value
32768
normalized value
32768
scope
source-model
source id
jina
confidence
1

jina ↗

field
embedding_dimensions
value
2048
normalized value
2048
scope
source-model
source id
jina
confidence
1

Original sources

1–6 / 6

Jina AI · Model page ↗

Hugging Face · jinaai/jina-embeddings-v4 ↗

arXiv · Paper · 2506.18902 ↗

Jina AI · Source snapshot ↗

Hugging Face · Catalog ↗

Jina AI · Catalog ↗

Source links

1–3 / 3

Explore more models

Models

Loading
EVIDENCE / COMPARISON

Benchmarks

Explore every benchmark

Loading

These charts contain model-linked results. The charts above also include unlinked source profiles. Compare scores within each evaluation. Model quality, usage, transparency and system performance measure different things.

Install TensorTrace