e5-large-v2

intfloat MIT out 3 years ago

335M parameters, text in, embeddings out, 512 tokens of context. 2 providers serve it from $0.010 per million input tokens.

Verified 2026-10-25 Rank #21 in Embeddings · score 98 method v1 Sources 3 Downloads 30d 851.9k · #126 of 619 tracked

Can I run it?0.2GB at INT4Estimatefits a 16 GB cardmethod

What does it cost?$0.010/1M incheapest of 1DeepInfra2026-10-24

Can I use it?MITPermissivelicence file2026-10-24

Does it work?62.79on MTEB#21 of 64 measuredleaderboard2026-10-24

Do people use it?851.9kdownloads, 30 days#126 of 619 we track · top 20%+2 likes, downloads −17% this week293 likesHugging Face2026-10-24

Worth reading

Picked by rule from 4 tracked pieces: the lab's own post, the most shared, a field report, the latest.

Field report

Embedding paragraphs from my blog with E5-large-v2

E5-large-v2 is an embedding model trained to match queries against text passages.

The embedding command stores the original text content in the database alongside the calculated vectors.
The author used a specific SQL query to extract paragraph content by stripping HTML tags before embedding.

til.simonwillison.net 2.3k words

4 best of 4 tracked

The 1 pick above, plus these 3. Top is the radar's score: discussion, source authority and the lab's own posts, fading with age.

ArticleInstructor-XL vs E5-large-v2: Embedding Model Guide | Inference Systemsinferensys.com · undated
Articlee5-large-v2 by Liang Wang: Open-Weight Modelsavrn.com · undated
ArticleE5-Large v2 | VIPS Learnlearn.engineering.vips.edu · 2026-04-20

Where to run it

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

#	Cheapest 2	In /1M	Out /1M	Context	Quant	Seen
1	DeepInfra	$0.010	—	512	—	Oct 24
2	HF Inference	—	—	—	—	Oct 25

All 2 providers →

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 2 API providers; 12 quantized builds for local use.

As a service2DeepInfra, HF Inference
On own hardware12quantized builds
Built on3634 fine-tunes · 2 adapters · 0 merges
In apps100+Hugging Face Spaces

Top Spaces running it: camel-ai/Paper2Poster ♥ 28 · agora-protocol/agora-demo ♥ 15 · Mqleet/AutoPage ♥ 12 · Pixeltable/Call-Analysis-AI-Tool ♥ 6

Will it fit?

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

Fits a 16 GB card at INT4

About 0.2 GB at INT4, 0.4 GB at FP8, 0.8 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.

#21 of 64 on MTEB so far

62.79. 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.

1 dated article and 15 repo commits so far

Teal is the weights (2023-05-19), amber triangles are 1 dated article (each links to it), grey ticks are repo commits.

What people built on it

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

48 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 ChristianAzinn/e5-large-v2-gguf for llama.cpp, Ollama, LM Studio 7.7k 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	49.31 nDCG@10 x100 (mean of 10 tasks)	10 tasks, split=test	leaderboard	2026-10-24
measured MTEB · Embed text for search	62.79 mean(task) x100 over 41 tasks	41 tasks, split=test	leaderboard	2026-10-24
measured MTEB Reranking · Rerank search results	45.72 MAP x100 (mean of 2 tasks)	2 tasks, split=test	leaderboard	2026-10-24

What the card says

1 cited sentence

Summary generated, cited

The architecture consists of 24 layers and an embedding size of 1024.^[1]

“This model has 24 layers and the embedding size is 1024.”

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

2 use cases

What it is good at 2 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

21 fields

Listed in Embeddings

Facts

e5-large-v2 has 335M parameters. (safetensors metadata, observed 2026-10-24)
Context length is 512 tokens (512). (config.json max_position_embeddings, observed 2026-10-24)
License: MIT (friction: none). Permissive. (license file, observed 2026-10-24)
Created 2023-05-19, last modified 2025-02-17. (HF API, observed 2026-10-24)

Pipeline tag	sentence-similarity
Library	sentence-transformers
Architectures	BertModel
Model type	bert
Dtype	float32
Params by dtype	F32: 335,141,888 · I64: 512
Size band	S
MoE	no
Context length	512 (config.json max_position_embeddings)
Languages	en
Papers	arXiv:2212.03533, arXiv:2104.08663, arXiv:2210.07316
License	MIT (mit)
Hub snapshot	2026-10-25

Servability

vLLM · SGLang

Servability

vLLM: supported · BertModel · 7050ef881ff0835f6890a47db1859a24c65481d6SGLang: supported · BertModel · 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 e5-large-v2: params × bytes/param × 1.2 overhead. KV cache and activations are extra. How this is computed.

BF16 (2 bytes/param)	0.8 GB · Edge (<4 GB)
FP8 (1 byte/param)	0.4 GB · Edge (<4 GB)
INT4 (0.5 bytes/param)	0.2 GB · Edge (<4 GB)

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

All variants

40 variants

Lineage

Original model Family: intfloat/e5-large-v2 48 derivatives on the Hub

finetune: 34quantized: 12adapter: 2

Top variants (40 shown, ranked by downloads)

Variant	Relation	Quant	Params	Band	License	Author	Likes	Downloads
OpenMed/OpenMed-PII-Italian-ClinicalE5-Large-335M-v1	finetune	—	334M	S	apache-2.0	OpenMed	0	107.6k
OpenMed/OpenMed-PII-ClinicalE5-Large-335M-v1	finetune	—	334M	S	apache-2.0	OpenMed	0	17.3k
Xenova/e5-large-v2	quantized	—	—	—	—	Xenova	6	15.5k
ChristianAzinn/e5-large-v2-gguf	quantized	gguf	—	—	mit	ChristianAzinn	2	7.7k
mradermacher/e5-large-v2-i1-GGUF	quantized	GGUF	—	—	mit	mradermacher	0	4.7k
mradermacher/e5-large-v2-GGUF	quantized	GGUF	—	—	mit	mradermacher	0	2.2k
mjwong/e5-large-v2-mnli	finetune	—	335M	S	mit	mjwong	1	1.1k
mangopy/ToolRet-trained-e5-large-v2	finetune	—	335M	S	—	mangopy	0	768
idanylenko/e5-large-v2-ctx1024	finetune	—	336M	S	mit	idanylenko	1	625
Hanno-Labs/tagline-quality-e5-ranker	quantized	—	335M	S	mit	Hanno-Labs	0	586
clips/e5-large-v2-t2t	finetune	—	335M	S	mit	clips	0	518
dostoevskyIdiot/intfloat-e5-large-v2-jaiv-v2	finetune	—	335M	S	—	dostoevskyIdiot	0	395
juanmcristobal/e5-large-v2-mitre	finetune	—	335M	S	—	juanmcristobal	1	338
hongming/e5-large-v2-nli-v1	finetune	—	335M	S	—	hongming	0	363
xplainlp/e5-large-v2-climatecheck	finetune	—	335M	S	—	xplainlp	0	356
ainativestudio/locomo-e5-large-finetuned	finetune	—	335M	S	—	ainativestudio	0	355
noystl/recomb-pred-e5-large	finetune	—	335M	S	cc	noystl	0	338
Pekarnick/e5-large-v2-Q4_K_M-GGUF	quantized	GGUF	—	—	mit	Pekarnick	0	316
craigtrim/edu-alias-mapper-e5	finetune	—	335M	S	—	craigtrim	0	200
OpenMed/OpenMed-PII-German-ClinicalE5-Large-335M-v1	finetune	—	334M	S	apache-2.0	OpenMed	0	178
MinhPhuc0804/e5-docling-checkthat-task1-v1	finetune	—	335M	S	—	MinhPhuc0804	0	160
OpenMed/OpenMed-PII-French-ClinicalE5-Large-335M-v1	finetune	—	334M	S	apache-2.0	OpenMed	0	158
OpenMed/OpenMed-PII-Dutch-ClinicalE5-Large-335M-v1	finetune	—	334M	S	apache-2.0	OpenMed	0	143
petkopetkov/e5-large-v2-patent	finetune	—	335M	S	—	petkopetkov	0	139
OpenMed/OpenMed-PII-Spanish-ClinicalE5-Large-335M-v1	finetune	—	334M	S	apache-2.0	OpenMed	0	126
IshTale/EccomerceEmbeddingModel	finetune	—	335M	S	—	IshTale	0	106
samanvitha7/semeval2026-e5_large-checkpoints	finetune	—	335M	S	—	samanvitha7	0	105
OpenMed/OpenMed-PII-Portuguese-ClinicalE5-Large-335M-v1	finetune	—	334M	S	apache-2.0	OpenMed	0	97
OpenMed/OpenMed-PII-Telugu-ClinicalE5-Large-335M-v1	finetune	—	334M	S	apache-2.0	OpenMed	0	92
OpenMed/OpenMed-PII-Turkish-ClinicalE5-Large-335M-v1	finetune	—	334M	S	apache-2.0	OpenMed	0	89
jrc2139/e5-large-v2-ONNX	quantized	—	—	—	—	jrc2139	0	85
OpenMed/OpenMed-PII-Hindi-ClinicalE5-Large-335M-v1	finetune	—	334M	S	apache-2.0	OpenMed	0	85
keisuke-miyako/e5-large-v2-gguf-q4_k_m	quantized	gguf	—	—	—	keisuke-miyako	0	76
keisuke-miyako/e5-large-v2-onnx-f32	quantized	—	—	—	—	keisuke-miyako	0	65
keisuke-miyako/e5-large-v2-gguf-q8_0	quantized	gguf	—	—	—	keisuke-miyako	0	65
keisuke-miyako/e5-large-v2-ct2-float16	finetune	—	—	—	—	keisuke-miyako	0	51
keisuke-miyako/e5-large-v2-ct2-int8_float16	finetune	—	—	—	—	keisuke-miyako	0	50
richardyoung/CardioEmbed-E5-large-v2	adapter	—	—	—	apache-2.0	richardyoung	0	42
keisuke-miyako/e5-large-v2-onnx-f16	quantized	—	—	—	—	keisuke-miyako	0	38
abdulrahman-nuzha/intfloat-e5-large-v2-english-fp16-allagree	finetune	-fp16	335M	S	mit	abdulrahman-nuzha	0	36

Repository changelog

15 commits

Repository changelog 15 commits

Date	Commit	SHA
2025-02-17	Add exported openvino model 'openvino_model_qint8_quantized.xml' (#20)	f169b11
2025-02-17	Add exported openvino model 'openvino_model.xml' (#19)	756b8dd
2025-02-17	Add exported onnx model 'model_O4.onnx' (#18)	226d082
2025-02-17	Add exported onnx model 'model_qint8_avx512_vnni.onnx' (#17)	febed3b
2025-02-17	Add exported onnx model 'model.onnx' (#16)	ef82eb1
2023-08-07	Update README.md	b322e09
2023-07-27	add sentence similarity tag	86377ff
2023-07-20	Update README.md	21903a7
2023-07-01	create handler.py (#8)	3bd6250
2023-06-20	Adding `safetensors` variant of this model (#7)	b303079
2023-06-18	sentence transformers compat (#6)	cc7480b
2023-05-22	Update README.md	d949600
2023-05-19	Update README.md	022c5a7
2023-05-19	upload model weights	bdaaca6
2023-05-19	initial commit	276398b

All providers

2 providers · 2 endpoints

Where to run 2 providers · 2 endpoints

Hosted endpoints observed for e5-large-v2, 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: $0.01/M.

Provider	Model ref	Context	Quant	$/M in	$/M out	tok/s	TTFT	Tools	Status	Observed
DeepInfra deepinfra	intfloat/e5-large-v2	512	—	$0.01	—	—	—	—	live	2026-10-24
HF Inference hf	intfloat/e5-large-v2	—	—	—	—	—	—	—	live	2026-10-25

Articles and coverage

4 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.

Analysis4

Embedding paragraphs from my blog with E5-large-v2

til.simonwillison.netauthority2,297 words

E5-large-v2 is an embedding model trained to match queries against text passages. The author used the llm tool with the llm-sentence-transformers plugin to generate embeddings for 18,918 blog paragraphs, downloading 2.5GB of model files. The model truncates text at 512 tokens, requires 'passage:' and 'query:' prefixes, and the process took about 25 minutes.written, reviewed, cited

The embedding command stores the original text content in the database alongside the calculated vectors.
The author used a specific SQL query to extract paragraph content by stripping HTML tags before embedding.

Instructor-XL vs E5-large-v2: Embedding Model Guide | Inference Systems

inferensys.com1,249 words

Instructor-XL adapts to specific retrieval tasks using natural language instructions, achieving a 3-5 point nDCG@10 improvement on asymmetric BEIR benchmarks like NFCorpus compared to generic prompts. The 1.5B parameter model supports 768-dimensional embeddings and requires careful instruction engineering, whereas the 335M parameter E5-large-v2 offers ~2,000 sentences/sec inference speed with 1024-dimensional embeddings for general-purpose retrieval. Teams must weigh the operational overhead of prompt templates against the measurable precision boost for specialized domains like legal document review.written, reviewed, cited

Instructor-XL supports over 300 instruction templates to cover diverse use cases without fine-tuning.

e5-large-v2 by Liang Wang: Open-Weight Model

savrn.com2,574 words

e5-large-v2 is an open-weight sentence similarity model by Liang Wang released under the MIT License. It features 335M parameters, 24 layers, and a 1024 embedding size with a 512-token context. The 6.7 GB repository draws 1M downloads a month and requires about 0.8 GB of GPU memory at 16-bit precision. Released in May 2023, it fits on 1x MI300X from $1.85 an hour.written, reviewed, cited

The repository is 6.7 GB because weights are stored in float32 and shipped in four formats.
Engineers must add query and passage prefixes to inputs before benchmarking the model.
The 192 GB card capacity is far more than a 335M-parameter BERT encoder actually needs.

E5-Large v2 | VIPS Learn

learn.engineering.vips.edu2026-04-20303 words

Microsoft released E5-Large v2 in December 2022 as a 335-million-parameter text embedding model trained with the E5 contrastive recipe. The model operates under MIT-style licensing but is limited by a 512-token input window that restricts long-document embedding capabilities. Engineers can download weights from Hugging Face under 'intfloat/e5-large-v2' for affordable RAG pipelines on CPU or small GPUs.written, reviewed, cited

Papers

3 papers

Papers 3

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).

MTEB: Massive Text Embedding Benchmark cited in card

Niklas Muennighoff, Nouamane Tazi, Loïc Magne +1 · 2022-10-13 · PDF · HF paper page (9 upvotes)
BEIR: A Heterogenous Benchmark for Zero-shot Evaluation of Information Retrieval Models cited in card

Nandan Thakur, Nils Reimers, Andreas Rücklé +2 · 2021-04-17 · PDF · HF paper page (5 upvotes)
Text Embeddings by Weakly-Supervised Contrastive Pre-training cited in card

Liang Wang, Nan Yang, Xiaolong Huang +5 · 2022-12-07 · PDF · HF paper page (2 upvotes)

Similar models

3 models

Similar models

embeddinggemma-300m · 303M
same size band (S), same task (sentence-similarity)
gte-large · 335M
same size band (S), same task (sentence-similarity), permissive license
all-roberta-large-v1 · 355M
same size band (S), same task (sentence-similarity), fewer 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 e5-large-v2

Watch this model

/feeds/models/intfloat--e5-large-v2.xml(RSS 2.0) · every change to e5-large-v2
/feeds/families/intfloat--e5-large-v2.xml(RSS 2.0) · every change across the e5-large-v2 family

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