jina-embeddings-v3

572M parameters, text in, embeddings out, 8K tokens of context.

Verified 2026-10-25 Rank #26 in Embeddings · score 96.3 method v1 Sources 3 Downloads 30d 1.6M · #77 of 619 tracked

Can I run it?0.3GB at INT4Estimatefits a 16 GB cardmethod

Can I use it?CC BY-NC 4.0Non-commercial only; commercial hosting not permitted without a separate licenselicence file2026-10-24

Does it work?66.6on MTEB#11 of 64 measuredleaderboard2026-10-24

Do people use it?1.6Mdownloads, 30 days#77 of 619 we track · top 12%+3 likes, downloads −8% this week1,171 likesHugging Face2026-10-24

Worth reading

Picked by rule from 8 tracked pieces: the lab's own post, the most shared, a field report, the latest.

The lab's own post

Jina Embeddings V3: A Frontier Multilingual Embedding Model

Jina Embeddings v3 is a 570 million parameter model released on September 18, 2024, that supports inputs up to 8192 tokens and outputs 1024 dimensions.

The model inserts task-specific LoRA adapters into every transformer layer and performs encoding in one shot.
When late_chunking is enabled, the total token count per request is restricted to 8192 regardless of the number of sentences.

jina.ai 2024-09-18 3 HN pts 1.4k words

8 best of 8 tracked

The 2 picks above, plus these 6. Top is the radar's score: discussion, source authority and the lab's own posts, fading with age.

Article[Literature Review] jina-embeddings-v3: Multilingual Embeddings With Task LoRAthemoonlight.io · undated
ArticleJina Embeddings v3 | OSS AI Hubossaihub.com · undated
ArticlePapers Explained 266: Jina Embeddings v3 | by Ritvik Rastogi | Mediumritvik19.medium.com · 2024-12-05
Articlejina-embeddings-v3: Multilingual Embeddings With Task LoRA | alphaXivalphaxiv.org · 2024-09-19
Articlejina-embeddings-v3: Multilingual Embeddings With Task LoRAexa.ai · 2024-09-16
ArticleJina Embeddings V3: Multilingual Task LoRAemergentmind.com · 2024-09-16

Show all 6

Where it's used

Hugging Face does not say who downloads a model or with what. These are the proxies: API providers listing it, quantized builds, fine-tunes/adapters/merges, and Spaces that load it.

No API provider serves it, so its downloads are people running it themselves: 5 quantized builds and 67 Spaces.

As a service0no API provider listed
On own hardware5quantized builds
Built on2828 fine-tunes · 0 adapters · 0 merges
In apps67Hugging Face Spaces

Top Spaces running it: embedl/hfviewer ♥ 181 · oceansweep/tldw ♥ 27 · antoinelouis/mtem-pruner ♥ 14 · lihuigu/SciPIP ♥ 7

Will it fit?

Weights only, an estimate: leave room for the KV cache. Method

Fits a 16 GB card at INT4

About 0.3 GB at INT4, 0.7 GB at FP8, 1.4 GB at BF16: weights only, so leave room for the KV cache.

Estimate params × bytes per param × 1.2 · how we estimate

How it scores

Third-party measurements only; author-reported numbers stay in Reference, marked as such.

#11 of 64 on MTEB so far

66.6. Each tick is a model on this site with a measured MTEB row; author-reported numbers stay in Reference. Mean over 40/41 tasks present. 40/41 tasks

Source · leaderboard · 2026-10-24

The story so far

Weights, every dated article, and repo commits on one rail.

Most coverage came in the first two weeks

Teal is the weights (2024-09-05), amber triangles are 5 dated articles (each links to it), grey ticks are repo commits.

What people built on it

33 derivatives on the Hub. Which build to run, by runtime: the most downloaded in each class.

33 derivatives on the Hub, most of them fine-tunes

From the Hugging Face model tree; quantized builds in teal, fine-tunes, merges and adapters in greys.

GGUF second-state/jina-embeddings-v3-GGUF for llama.cpp, Ollama, LM Studio 13.3k downloads

All variants →

Reference

Everything else, folded: the full tables, every source.

Benchmarks

3 rows

Benchmarks 3 rows

measured third-party leaderboard row (3)author-reported number from the model card or eval filesNo confidence intervals unless the source gave them; settings differ between rows, so compare within a benchmark, not across. Date = the benchmark's own date when it gives one, else when we read it.

Benchmark	Value	Setting / notes	Reported by	Date
measured MTEB · Embed text for search	54.29 nDCG@10 x100 (mean of 10 tasks)	10 tasks, split=test	leaderboard	2026-10-24
measured MTEB · Embed text for search	66.6 mean over 40/41 tasks present	40/41 tasks	leaderboard	2026-10-24
measured MTEB Reranking · Rerank search results	47.94 MAP x100 (mean of 2 tasks)	2 tasks, split=test	leaderboard	2026-10-24

What the card says

2 cited sentences

Summary generated, cited

jina-embeddings-v3 is a multilingual multi-task text embedding model with 572M parameters.^[1]^[2] The architecture is based on XLM-RoBERTa and supports input sequences up to 8192 tokens.^[3]^[4]

parameters human
“multilingual multi-task text embedding model”
“Based on the Jina-XLM-RoBERTa architecture”
“up to 8192 tokens”

Generated 2026-10-07 15:01 UTC by parasail-gemma-4-26b-a4b-it; each numeral points at the fact it rests on. Not a verdict.

Use-case evidence

3 use cases

What it is good at 3 use cases

Labels are evidence, not verdicts: measured a third-party leaderboard row for this model; author-reported a number the model author published in the card or eval files; structural a deterministic signal from the name, config or chat template, or a decision-model estimate. Each chip links to its source. All use cases.

Facts

20 fields

Listed in Embeddings

Facts

jina-embeddings-v3 has 572M parameters. (safetensors metadata, observed 2026-10-24)
Context length is 8,194 tokens (8K). (config.json max_position_embeddings, observed 2026-10-24)
License: CC BY-NC 4.0 (friction: non-commercial). Non-commercial only; commercial hosting not permitted without a separate license. (license file, observed 2026-10-24)
Created 2024-09-05, last modified 2026-04-08. (HF API, observed 2026-10-24)

Pipeline tag	feature-extraction
Library	transformers
Architectures	XLMRobertaModel
Dtype	bfloat16
Params by dtype	BF16: 572,310,396
Size band	S
MoE	no
Context length	8,194 (config.json max_position_embeddings)
Languages	multilingual, af, am, ar, as, az, be, bg, bn, br, bs, ca, cs, cy, da, de, el, en, eo, es, et, eu, fa, fi, fr, fy, ga, gd, gl, gu, ha, he, hi, hr, hu, hy, id, is, it, ja, jv, ka, kk, km, kn, ko, ku, ky, la, lo, lt, lv, mg, mk, ml, mn, mr, ms, my, ne, nl, false, om, or, pa, pl, ps, pt, ro, ru, sa, sd, si, sk, sl, so, sq, sr, su, sv, sw, ta, te, th, tl, tr, ug, uk, ur, uz, vi, xh, yi, zh
Papers	arXiv:2409.10173
License	CC BY-NC 4.0 (cc-by-nc-4.0)
Hub snapshot	2026-10-25

Servability

vLLM · SGLang

Servability

vLLM: supported · XLMRobertaModel · 7050ef881ff0835f6890a47db1859a24c65481d6SGLang: supported · XLMRobertaModel · 773cb938fe9484a8f7e7e46fb1b1fe96412c59ed

vLLM registry (7050ef881ff0835f6890a47db1859a24c65481d6) · SGLang registry (773cb938fe9484a8f7e7e46fb1b1fe96412c59ed) · Matched by architecture class name against each project's registry on main; 'supported' means the architecture is registered, not that this checkpoint was tested.

Hardware fit

3 estimates

Hardware fit ESTIMATE

Weights-only memory for jina-embeddings-v3: params × bytes/param × 1.2 overhead. KV cache and activations are extra. How this is computed.

BF16 (2 bytes/param)	1.4 GB · Edge (<4 GB)
FP8 (1 byte/param)	0.7 GB · Edge (<4 GB)
INT4 (0.5 bytes/param)	0.3 GB · Edge (<4 GB)

Weights only (params × bytes/param × 1.2 overhead); KV cache and activations extra.

All variants

33 variants

Lineage

Original model Family: jinaai/jina-embeddings-v3 33 derivatives on the Hub

finetune: 28quantized: 5

Top variants (33 shown, ranked by downloads)

Variant	Relation	Quant	Params	Band	License	Author	Likes	Downloads
CISCai/jina-embeddings-v3-separation-distilled	finetune	—	64.0M	S	cc-by-nc-4.0	CISCai	2	189.1k
johnpaulbin/jina-embeddings-v3-128	finetune	—	32.0M	S	mit	johnpaulbin	0	26.3k
alikia2x/jina-embedding-v3-m2v-1024	quantized	—	256M	S	mit	alikia2x	5	23.5k
second-state/jina-embeddings-v3-GGUF	quantized	GGUF	—	—	—	second-state	3	13.3k
gaianet/jina-embeddings-v3-GGUF	quantized	GGUF	—	—	—	gaianet	1	8.2k
CISCai/jina-embeddings-v3-matching-distilled	finetune	—	64.0M	S	cc-by-nc-4.0	CISCai	0	5.6k
hs-hf/jina-embeddings-v3-distilled	finetune	—	128M	S	cc-by-nc-4.0	hs-hf	14	4.7k
CISCai/jina-embeddings-v3-classification-distilled	finetune	—	64.0M	S	cc-by-nc-4.0	CISCai	0	2.1k
CISCai/jina-embeddings-v3-query-distilled	finetune	—	64.0M	S	cc-by-nc-4.0	CISCai	1	757
SajjadMajdi/jina-embedding-v3	finetune	—	572M	S	apache-2.0	SajjadMajdi	0	708
alikia2x/jina-embedding-v3-m2v-256	quantized	—	64.0M	S	mit	alikia2x	1	439
NAMAA-Space/zarra	finetune	—	64.0M	S	mit	NAMAA-Space	2	387
angelitasr/jina-embeddings-v3_eeid	finetune	—	572M	S	—	angelitasr	1	394
Jrinky/jina_final_temp	finetune	—	572M	S	—	Jrinky	0	419
kaleinaNyan/jina-v3-rullmarena-judge	finetune	—	614M	S	apache-2.0	kaleinaNyan	3	261
csanz91/lampistero_rag_embeddings	finetune	—	572M	S	apache-2.0	csanz91	1	259
kaleinaNyan/jina-v3-rullmarena-judge-300924	finetune	—	589M	S	apache-2.0	kaleinaNyan	2	154
CISCai/jina-embeddings-v3-passage-distilled	finetune	—	64.0M	S	cc-by-nc-4.0	CISCai	1	199
kaleinaNyan/jina-v3-rullmarena-judge-041024	finetune	—	589M	S	apache-2.0	kaleinaNyan	1	179
ldwformat/jina-embeddings-v3-Q8-onnx	quantized	—	—	—	—	ldwformat	0	189
yodyamahesa/paper-recommendation-jinav3-keyword	finetune	—	572M	S	—	yodyamahesa	0	166
Thaweewat/jina-embedding-v3-m2v-128	finetune	—	—	—	—	Thaweewat	3	0
csanz91/lampistero_rag_embeddings_2	finetune	—	572M	S	apache-2.0	csanz91	0	150
FinancialReports/jina-v3-financial-classifier	finetune	—	572M	S	apache-2.0	FinancialReports	0	134
Abdelkareem/jina_v3_distilled	finetune	—	64.0M	S	mit	Abdelkareem	0	73
tboquet/m2v-jina-embeddings-v3-pca-256	finetune	—	64.0M	S	mit	tboquet	0	53
Thaweewat/jina-embedding-v3-m2v-1024	finetune	—	—	—	—	Thaweewat	1	0
YeonwooSung/jina-embeddings-v3-msmarco-v3-bpr	finetune	—	—	—	—	YeonwooSung	1	0
Thaweewat/jina-embedding-v3-m2v-256	finetune	—	—	—	—	Thaweewat	0	0
Thaweewat/jina-embedding-v3-m2v-512	finetune	—	—	—	—	Thaweewat	0	0
Thaweewat/jina-embedding-v3-m2v-768	finetune	—	—	—	—	Thaweewat	0	0
Mercity/memory-retrieval-jina-v3-lora	finetune	—	—	—	—	Mercity	0	0
siilasgre99/jinamodel	finetune	—	—	—	—	siilasgre99	0	0

Repository changelog

30 commits

changed in the last 7 daysRank band moves: 2026-10-24all changes

Repository changelog 30 commits

Date	Commit	SHA
2026-04-08	Integrate with Sentence Transformers v5.4 (#139)	ab036b0
2025-02-24	feat-add-cmteb-qbqtc (#115)	f1944de
2025-01-06	chore-fix-logo (#104)	2893bc1
2024-12-03	readme: azure link (#93)	62a8174
2024-11-27	fix: normalization after truncation (#83)	30996fe
2024-11-24	readme: add onnx mean pool function (#82)	9c9e6d9
2024-11-11	readme: onnx final outputs (#68)	c445d96
2024-10-23	Update custom_st.py to be compatible with Python 3.8 (#55)	fa78e35
2024-10-22	remove-st-prompt-length (#53)	80ffb06
2024-10-21	Save Transformer module in root when training with ST (#50)	215a6e1
2024-10-11	Add compatibility with Sentence Transformers v3.2.0 (#45)	4be32c2
2024-10-10	readme: finetuning (#43)	837f25a
2024-10-10	feat: finetuning via ST (#42)	ebf2504
2024-10-04	Update config.json	343dbf5
2024-09-27	Add Sentence Transformers tag (#37)	0959ff1
2024-09-24	Enable saving in SentenceTransformers by adding get_config_dict (#32)	54da5c0
2024-09-21	chore: patch readme on api	120c5ad
2024-09-21	chore: patch readme on api	685c080
2024-09-19	add polish retrieval socre (#29)	c4cb35c
2024-09-19	Update ONNX weights (#24)	7550804
2024-09-18	chore: patch readme license	921a3d2
2024-09-18	chore: patch readme license	7d49abc
2024-09-18	chore: patch multi-class classification rus scores	a408c93
2024-09-18	chore: patch polish ppc pc	58c59c2
2024-09-18	Merge branch 'main' of https://huggingface.co/jinaai/jina-embeddings-v2	0a6c97d
2024-09-18	chore: patch eighttags clustering pl	fee033e
2024-09-18	support ST for python3.8+ (#23)	e8fc0c7
2024-09-18	chore: patch cmedqa name	2a44e8e
2024-09-18	chore: patch xqpa fr	0602367
2024-09-17	chore: patch mindsmall reranking score	8676923

Articles and coverage

8 pieces

Ranked by source signals (HN, Lobsters, authority domains, official posts) and recency. Summaries are generated from a stored copy of the article text; every sentence cites a span of that text, and nothing is shown that failed the citation check.

Official1

Jina Embeddings V3: A Frontier Multilingual Embedding Model

jina.ai2024-09-18officialHN 3 pts · 0 comments1,437 words

Jina Embeddings v3 is a 570 million parameter model released on September 18, 2024, that supports inputs up to 8192 tokens and outputs 1024 dimensions. The model outperforms proprietary embeddings from OpenAI and Cohere on English tasks while surpassing multilingual-e5-large-instruct across all multilingual benchmarks. Users can truncate dimensions down to 32 without sacrificing performance, and the model supports 89 languages under a CC BY-NC 4.0 license.written, reviewed, cited

The model inserts task-specific LoRA adapters into every transformer layer and performs encoding in one shot.
When late_chunking is enabled, the total token count per request is restricted to 8192 regardless of the number of sentences.
The model is licensed under CC BY-NC 4.0, requiring contact for any commercial usage inquiries.

Analysis7

Jina Embeddings V3: Multilingual Text Encoder with Low-Rank Adaptations | Springer Nature Link

link.springer.com

[Literature Review] jina-embeddings-v3: Multilingual Embeddings With Task LoRA

themoonlight.io

Jina Embeddings v3 | OSS AI Hub

ossaihub.com2,887 words

Jina Embeddings v3 supports an 8192-token native context window and achieves an average MTEB score of ~54.3 under an Apache 2.0 license. The model uses task-specific LoRA adapters with less than 2% parameter overhead to handle retrieval, classification, and semantic similarity on a single checkpoint. On an NVIDIA A10G, it delivers 1,400–1,800 tokens/second throughput for 512-token inputs but drops to ~180–250 tokens/second at full context length.written, reviewed, cited

jina-embeddings-v3: Multilingual Embeddings With Task LoRA

exa.ai2024-09-16137 words

Jina Embeddings V3: Multilingual Task LoRA

emergentmind.com2024-09-16802 words

jina-embeddings-v3: Multilingual Embeddings With Task LoRA | alphaXiv

alphaxiv.org2024-09-191,280 words

Papers Explained 266: Jina Embeddings v3 | by Ritvik Rastogi | Medium

ritvik19.medium.com2024-12-05

Papers

1 paper

Papers 1

arXiv links found in the model card and its tags. "Peer-reviewed" is shown only with evidence: a published venue in OpenAlex (measured) or the authors' own journal reference on arXiv (reported).

jina-embeddings-v3: Multilingual Embeddings With Task LoRA this model

Saba Sturua, Isabelle Mohr, Mohammad Kalim Akram +8 · 2024-09-16 · PDF · HF paper page (37 upvotes)

Similar models

3 models

Similar models

multilingual-e5-large · 560M
same size band (S), same task (feature-extraction), more providers, permissive license
Qwen3-Embedding-0.6B · 596M
same size band (S), same task (feature-extraction), more providers, permissive license
multilingual-e5-large-instruct · 560M
same size band (S), same task (feature-extraction), more providers, permissive license

Sources

3 sources

Sources

Hub snapshot: 2026-10-25 · HF API: 2026-10-24 21:34 UTC · Providers: 2026-10-24 21:44 UTC · Articles: 2026-10-24 21:45 UTC · Summary: 2026-10-07 15:01 UTC · Last verified 2026-10-25

Follow jina-embeddings-v3

Watch this model

/feeds/models/jinaai--jina-embeddings-v3.xml(RSS 2.0) · every change to jina-embeddings-v3
/feeds/families/jinaai--jina-embeddings-v3.xml(RSS 2.0) · every change across the jina-embeddings-v3 family

Email alerts are not open yet; the feeds above are the same data. See feeds for the item schema.