Qwen3-Embedding-4B

4.02B parameters, text in, embeddings out, 40K tokens of context. 1 provider serves it from $0.020 per million input tokens.

Verified 2026-10-25 Rank #28 in Embeddings · score 95.4 method v1 Sources 3 Downloads 30d 1.8M · #67 of 619 tracked

Changed this week: 4 new quantized builds on the Hub (53 in all).2026-10-24 · 3 new quantized builds on the Hub (52 in all).2026-10-24

Can I run it?2.4GB at INT4Estimatefits a 16 GB cardmethod

What does it cost?$0.020/1M incheapest of 1DeepInfra2026-10-25

Can I use it?Apache 2.0Permissive; patent grantlicence file2026-10-24

Does it work?74.6on MTEB#3 of 64 measuredleaderboard2026-10-24

Do people use it?1.8Mdownloads, 30 days#67 of 619 we track · top 11%+2 likes, downloads +5% this week342 likesHugging Face2026-10-24

Worth reading

Picked by rule from 10 tracked pieces: the lab's own post, the most shared, a field report, the latest.

Latest · 8 months ago

Fine-Tuning Qwen3 Embeddings for product category classification

blog.ivan.digital 2026-03-03 2 HN pts about the Qwen3 Embedding family family

10 best of 10 tracked

The 1 pick above, plus these 9. Top is the radar's score: discussion, source authority and the lab's own posts, fading with age.

PaperQwen3 Embedding: Advancing Text ...arxiv.org · undated
ArticleQwen3 Embedding:Advancing Text Embedding and Reranking Through Foundation Modelsqwenlm.github.io · 2025-06-05 · 3 HN pts · official
PaperQwen3 Embedding: Advancing Text Embedding and Reranking with Foundation Modelsarxiv.org · 2025-06-11 · 1 HN pts
ArticleHands-on RAG with Qwen3 Embedding and Reranking ...milvus.io · undated
ArticleHow Qwen3 Embedding Beat Google at Its Own RAG Gamepub.towardsai.net · undated
ArticleQwen3 embeddings and rerankers: 0.6B vs 4B, text vs VL, ...superlinked.com · undated
CodeQwenLM/Qwen3-Embeddinggithub.com · 2025-06-05
ArticleQwen3 Embedding: Advancing Text Embedding and Reranking Through Foundation Models — Zen Blogzenlm.org · 2025-06-04
ArticleQwen3 Embedding 4B by Qwen | The GTM Directorythegtmdirectory.com · 2025-06-03

Show all 9

Where to run it

Live hosted endpoints, cheapest first, USD per million tokens. Ties break on output price, then provider name.

#	Cheapest live row	In /1M	Out /1M	Context	Quant	Seen
1	DeepInfra	$0.020	—	32K	—	Oct 25

All 1 provider →

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.

Served by 1 API provider; heavily run locally (53 quantized builds); a big base for builders (115 fine-tunes, adapters and merges).

As a service1DeepInfra
On own hardware53quantized builds
Built on11574 fine-tunes · 41 adapters · 0 merges
In apps26Hugging Face Spaces

Top Spaces running it: embedl/hfviewer ♥ 181 · ipepe/nomic-embeddings ♥ 2 · fx-tourist/qwen3-embedding-4b ♥ 2 · Abdulatif-ai/1-openai-help-docs-ai-agent ♥ 1

Will it fit?

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

Fits a 16 GB card at INT4

About 2.4 GB at INT4, 4.8 GB at FP8, 9.7 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.

#3 of 64 on MTEB so far

74.6. Each tick is a model on this site with a measured MTEB row; author-reported numbers stay in Reference. Mean(task) x100 over 41 tasks. 41 tasks, split=test

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 (2025-06-03), amber triangles are 6 dated articles (each links to it), grey ticks are repo commits.

What people built on it

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

168 derivatives on the Hub, led by 74 fine-tunes

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

GGUF mradermacher/Qwen3-Embedding-4B-i1-GGUF for llama.cpp, Ollama, LM Studio 7.9k downloads MLX mlx-community/Qwen3-Embedding-4B-4bit-DWQ for Apple silicon 62.4k downloads AWQ drawais/Qwen3-Embedding-4B-AWQ-INT4 for vLLM, SGLang 92.4k 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	68.46 nDCG@10 x100 (mean of 10 tasks)	10 tasks, split=test	leaderboard	2026-10-24
measured MTEB · Embed text for search	74.6 mean(task) x100 over 41 tasks	41 tasks, split=test	leaderboard	2026-10-24
measured MTEB Reranking · Rerank search results	50.76 MAP x100 (mean of 2 tasks)	2 tasks, split=test	leaderboard	2026-10-24

What the card says

2 cited sentences

Summary generated, cited

The model is designed for tasks such as text retrieval and classification, supporting a context length of 32k.^[1] It is released under the Apache 2.0 license and supports over 100 languages.^[2]^[3]

“text retrieval, code retrieval, text classification, text…”
license
“Supported Languages: 100+ Languages”

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

5 use cases

What it is good at 5 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

Qwen3-Embedding-4B has 4.02B parameters. (safetensors metadata, observed 2026-10-24)
Context length is 40,960 tokens (40K). (config.json max_position_embeddings, observed 2026-10-24)
License: Apache 2.0 (friction: none). Permissive; patent grant. (license file, observed 2026-10-24)
Created 2025-06-03, last modified 2025-06-20. (HF API, observed 2026-10-24)

Pipeline tag	feature-extraction
Library	sentence-transformers
Architectures	Qwen3ForCausalLM
Model type	qwen3
Dtype	bfloat16
Params by dtype	BF16: 4,021,774,336
Size band	M
MoE	no
Context length	40,960 (config.json max_position_embeddings)
Papers	arXiv:2506.05176
License	Apache 2.0 (apache-2.0)
Hub snapshot	2026-10-25

Servability

vLLM · SGLang

Servability

vLLM: supported · Qwen3ForCausalLM · 7050ef881ff0835f6890a47db1859a24c65481d6SGLang: supported · Qwen3ForCausalLM · 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 Qwen3-Embedding-4B: params × bytes/param × 1.2 overhead. KV cache and activations are extra. How this is computed.

BF16 (2 bytes/param)	9.7 GB · Single GPU (<24 GB)
FP8 (1 byte/param)	4.8 GB · Single GPU (<24 GB)
INT4 (0.5 bytes/param)	2.4 GB · Edge (<4 GB)

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

All variants

40 variants

Lineage

Fine-tuned from Qwen/Qwen3-4B-Base (same organization) Family: Qwen/Qwen3-Embedding 168 derivatives on the Hub

Base model: Qwen/Qwen3-4B-Base

quantized: 53finetune: 74adapter: 41

Top variants (40 shown, ranked by downloads)

Variant	Relation	Quant	Params	Band	License	Author	Likes	Downloads
boboliu/Qwen3-Embedding-4B-W4A16-G128	quantized	-W4A16	4.05B	M	apache-2.0	boboliu	5	2.5M
farbodtavakkoli/OTel-Embedding-4B	finetune	—	4.02B	M	apache-2.0	farbodtavakkoli	0	336.1k
Octen/Octen-Embedding-4B	finetune	—	4.02B	M	apache-2.0	Octen	28	289.7k
drawais/Qwen3-Embedding-4B-AWQ-INT4	quantized	AWQ	4.05B	M	apache-2.0	drawais	5	92.4k
mlx-community/Qwen3-Embedding-4B-4bit-DWQ	quantized	-4bit	4.02B	M	apache-2.0	mlx-community	19	62.4k
AQ-MedAI/Diver-Retriever-4B	finetune	—	4.02B	M	apache-2.0	AQ-MedAI	24	21k
ufca-llms/jua-4B-mixed	finetune	—	4.02B	M	mit	ufca-llms	2	15.5k
AQ-MedAI/Diver-Retriever-4B-1020	finetune	—	4.02B	M	apache-2.0	AQ-MedAI	6	11.3k
furiosa-ai/Qwen3-Embedding-4B	finetune	—	4.02B	M	apache-2.0	furiosa-ai	0	9.9k
Tevatron/AgentIR-4B	finetune	—	4.02B	M	—	Tevatron	9	9k
mradermacher/Qwen3-Embedding-4B-i1-GGUF	quantized	GGUF	—	—	apache-2.0	mradermacher	1	7.9k
Octen/Octen-Embedding-4B-INT8	quantized	INT8	4.02B	M	apache-2.0	Octen	6	7.4k
mradermacher/Qwen3-Embedding-4B-GGUF	quantized	GGUF	—	—	apache-2.0	mradermacher	1	6.8k
itsbohara/qwen3_embedding_4b_q8_0	quantized	GGUF	—	—	apache-2.0	itsbohara	6	6.4k
Casual-Autopsy/Qwen3-Embedding-4B-GGUFs	quantized	GGUF	—	—	apache-2.0	Casual-Autopsy	1	5.8k
enacimie/Qwen3-Embedding-4B-Q4_K_M-GGUF	quantized	GGUF	—	—	apache-2.0	enacimie	4	4.4k
DCC-BS/Qwen3-Embedding-4B-FP8-Dynamic	quantized	-FP8	4.02B	M	—	DCC-BS	1	4.1k
Querit/Querit-4B	finetune	—	201K	S	apache-2.0	Querit	6	3.6k
bflhc/MoD-Embedding	finetune	—	4.02B	M	apache-2.0	bflhc	3	3.6k
yale-nlp/RTriever-4B	finetune	—	4.02B	M	mit	yale-nlp	2	2.8k
lainlives/Qwen3-Embedding-4B-bnb-4bit	quantized	-bnb	4.14B	M	apache-2.0	lainlives	0	2.5k
cstr/qwen3-embed-4b-GGUF	quantized	GGUF	—	—	apache-2.0	cstr	0	2.5k
dolfsai/Qwen3-Embedding-4B-W4A16_ASYM	quantized	-W4A16	4.47B	M	apache-2.0	dolfsai	0	2.1k
batiai/Qwen3-Embedding-4B-GGUF	quantized	GGUF	—	—	apache-2.0	batiai	0	1.8k
QuaduxIT/Qwen3-Embedding-4B-W8A16	quantized	—	4.02B	M	apache-2.0	QuaduxIT	0	1.6k
iamthecage/Qwen3-Embedding-4B-OptiQ-4bit	quantized	-4bit	4.02B	M	apache-2.0	iamthecage	0	1.6k
shawnw3i/Qwen3-Embedding-4B-ONNX	quantized	—	—	—	apache-2.0	shawnw3i	0	1.4k
majentik/Qwen3-Embedding-4B-GGUF-Q8_0	quantized	GGUF	—	—	apache-2.0	majentik	2	1.3k
endyjasmi/Qwen3-Embedding-4B-Q4_K_M-GGUF	quantized	GGUF	—	—	apache-2.0	endyjasmi	0	1.3k
TakalaWang/qwen3-embedding-4B-code-search	finetune	—	4.02B	M	—	TakalaWang	1	1k
Hanno-Labs/dinghy-law-4b-v1	finetune	—	4.02B	M	apache-2.0	Hanno-Labs	2	942
majentik/Qwen3-Embedding-4B-MLX-8bit	finetune	MLX	4.02B	M	apache-2.0	majentik	0	966
chantomkit/qwen3-4B-risk-mnr-finance	finetune	—	4.02B	M	—	chantomkit	0	955
chantomkit/qwen3-4B-risk-mnr-education	finetune	—	4.02B	M	—	chantomkit	0	942
chantomkit/qwen3-4B-risk-mnr-law	finetune	—	4.02B	M	—	chantomkit	0	925
Geofront/Qwen3-Embedding-4B-GGUF	quantized	GGUF	—	—	apache-2.0	Geofront	0	907
chantomkit/qwen3-4B-risk-mnr-cyber	finetune	—	4.02B	M	—	chantomkit	0	875
fabricioalmeida/BumbaLM-Embedding-4B-v0.1	finetune	—	4.02B	M	—	fabricioalmeida	2	758
novacardsai/qwen3-medical	finetune	—	4.02B	M	—	novacardsai	0	826
miloK1/Qwen3-Embedding-4B-Q8_0-GGUF	quantized	Q8_0	—	—	apache-2.0	miloK1	1	706

Repository changelog

19 commits

changed in the last 7 daysOther: 2026-10-24, 2026-10-24all changes

Repository changelog 19 commits

Date	Commit	SHA
2025-06-20	Update README.md with TEI support (#13)	5cf2132
2025-06-07	add use case for vllm & modify Citation (#4)	408b81b
2025-06-06	Automatically add EOS via Tokenizer, integrate Sentence Transformers (#1)	636cd9b
2025-06-05	Update README.md	d69a721
2025-06-05	Update README.md	47a68a0
2025-06-05	Update README.md	afe8f7a
2025-06-05	Delete special_tokens_map.json	024c874
2025-06-05	Delete added_tokens.json	f4ff73b
2025-06-05	Update README.md	328a83c
2025-06-05	Update README.md	af551aa
2025-06-05	Update README.md	f28dd6d
2025-06-05	Update README.md	cb02917
2025-06-05	Update config.json	bcf11cc
2025-06-05	Delete modeling.py	1bf9dde
2025-06-03	Delete configuration.json	90818ca
2025-06-03	Delete .mv	04999d0
2025-06-03	Delete .msc	0a555bd
2025-06-03	Upload folder using huggingface_hub	d50085f
2025-06-03	initial commit	e55ad91

All providers

1 provider · 1 endpoint

Where to run 1 providers · 1 endpoints

Hosted endpoints observed for Qwen3-Embedding-4B, sorted by input price (USD per million tokens). Each row names its observer and date; prices change daily, so check the source before quoting. Cheapest observed input price: $0.02/M.

Provider	Model ref	Context	Quant	$/M in	$/M out	tok/s	TTFT	Tools	Status	Observed
DeepInfra hf, deepinfra	Qwen/Qwen3-Embedding-4B	32K	—	$0.02	—	—	—	—	live	2026-10-25

Articles and coverage

10 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

Qwen3 Embedding:Advancing Text Embedding and Reranking Through Foundation Models

qwenlm.github.io2025-06-05officialauthorityabout the Qwen3 Embedding familyHN 3 pts · 0 comments937 words

The Qwen3-Embedding-8B model ranks No.1 in the MTEB multilingual leaderboard with a score of 70.58. The Qwen3-Reranker-4B model achieves a FollowIR score of 14.84.generated, cited

Analysis8

Qwen3 Embedding: Advancing Text ...

arxiv.orgauthorityabout the Qwen3 Embedding family5,970 words

The Qwen3 Embedding series provides text embedding and reranking models in 0.6B, 4B, and 8B parameter sizes under the Apache 2.0 license. The flagship Qwen3-8B-Embedding model attains a score of 70.58 on the MTEB Multilingual benchmark and 80.68 on the MTEB Code benchmark. The Qwen3-Reranker-8B model improves ranking results by 3.0 points over the 0.6B model across multiple tasks.written, reviewed, cited

The Qwen3 Embedding series supports flexible dimension representation for embedding models.
Embedding models support customizing the input instruction according to different tasks.

Qwen3 Embedding: Advancing Text Embedding and Reranking with Foundation Models

arxiv.org2025-06-11authorityabout the Qwen3 Embedding familyHN 1 pts · 0 comments433 words

The Qwen3 Embedding series introduces new text embedding and reranking models built on Qwen3 foundation models. It offers a spectrum of model sizes (0.6B, 4B, 8B) for both embedding and reranking tasks and is publicly available under the Apache 2.0 license. The series achieves strong results on the multilingual evaluation benchmark MTEB for text embedding.written, reviewed, cited

Fine-Tuning Qwen3 Embeddings for product category classification

blog.ivan.digital2026-03-03about the Qwen3 Embedding familyHN 2 pts · 0 comments

Hands-on RAG with Qwen3 Embedding and Reranking ...

milvus.ioabout the Qwen3 Embedding family2,112 words

Alibaba released Qwen3 Embedding and Reranking models in three sizes (0.6B, 4B, 8B) designed for retrieval tasks. The Qwen3-Embedding-8B achieved a score of 70.58 on the MTEB multilingual leaderboard, surpassing BGE, E5, and Google Gemini. The Qwen3-Reranker-8B hit 69.02 on multilingual ranking tasks. These models support multilingual embeddings across 100+ languages and 32K context length.written, reviewed, cited

The embedding model uses a dual-encoder architecture while the reranker uses a cross-encoder setup.
Milvus Lite is suitable for local development, whereas production deployments require a server URI or cloud endpoint.

How Qwen3 Embedding Beat Google at Its Own RAG Game

pub.towardsai.netabout the Qwen3 Embedding family

QwenLM/Qwen3-Embedding

github.com2025-06-05authorityabout the Qwen3 Embedding family2,980 words

Qwen3 Embedding 4B by Qwen | The GTM Directory

thegtmdirectory.com2025-06-031,666 words

Qwen3 Embedding: Advancing Text Embedding and Reranking Through Foundation Models — Zen Blog

zenlm.org2025-06-04about the Qwen3 Embedding family964 words

All coverage tracked for this model (10) — append-only, nothing is ever removed

Title	Source	Published	First seen	Signals
Qwen3 Embedding: Advancing Text ...	arxiv.org	—	2026-10-07	exa
Hands-on RAG with Qwen3 Embedding and Reranking ...	milvus.io	—	2026-10-07	exa
How Qwen3 Embedding Beat Google at Its Own RAG Game	pub.towardsai.net	—	2026-10-07	exa
Qwen3 embeddings and rerankers: 0.6B vs 4B, text vs VL, ...	superlinked.com	—	2026-10-07	exa
Fine-Tuning Qwen3 Embeddings for product category classification	blog.ivan.digital	2026-03-03	2026-10-07	hn
Qwen3 Embedding: Advancing Text Embedding and Reranking with Foundation Models	arxiv.org	2025-06-11	2026-10-07	hn
Qwen3 Embedding:Advancing Text Embedding and Reranking Through Foundation Models	qwenlm.github.io official	2025-06-05	2026-10-07	hn
QwenLM/Qwen3-Embedding	github.com	2025-06-05	2026-10-07	exa
Qwen3 Embedding: Advancing Text Embedding and Reranking Through Foundation Models — Zen Blog	zenlm.org	2025-06-04	2026-10-07	exa
Qwen3 Embedding 4B by Qwen \| The GTM Directory	thegtmdirectory.com	2025-06-03	2026-10-07	exa

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).

Qwen3 Embedding: Advancing Text Embedding and Reranking Through Foundation Models this model

Yanzhao Zhang, Mingxin Li, Dingkun Long +9 · 2025-06-05 · PDF · HF paper page (86 upvotes)

Similar models

3 models

Similar models

bge-multilingual-gemma2 · 9.24B
same task (feature-extraction)
Qwen3-Reranker-4B · 4.02B
same size band (M), permissive license
bge-large-zh-v1.5
same task (feature-extraction), 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 Qwen3-Embedding-4B

Watch this model

/feeds/models/Qwen--Qwen3-Embedding-4B.xml(RSS 2.0) · every change to Qwen3-Embedding-4B
/feeds/families/Qwen--Qwen3-Embedding.xml(RSS 2.0) · every change across the Qwen3-Embedding family

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