gte-multilingual-base

305M parameters, text in, embeddings out, 8K tokens of context.

Verified 2026-10-25 Rank #44 in Embeddings · score 90.5 method v1 Sources 3 Downloads 30d 1M · #107 of 619 tracked

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

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

Does it work?62.51on MTEB#23 of 64 measuredleaderboard2026-10-24

Do people use it?1Mdownloads, 30 days#107 of 619 we track · top 17%+3 likes, downloads −21% this week390 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.

Latest · 6 months ago

Gte-go – my over-optimized embedding model

rcarmo.github.io 2026-04-27 4 HN pts · 1 comments about the gte family family

8 best of 8 tracked

The 1 pick above, plus these 7. Top is the radar's score: discussion, source authority and the lab's own posts, fading with age.

ArticleGTE-Multilingual Series: A Key Model for Retrieval- ...alibabacloud.com · undated · official
CodeRcarmo/gte-go: Golang inference for the GTE Small embedding modelgithub.com · 2026-04-25 · 1 HN pts
Articlegte-multilingual-base by Alibaba-NLP: Open-Weight Modelsavrn.com · undated
Articlegte-multilingual-base by Alibaba-NLP | Mixpeek Model Hubmixpeek.com · undated
ArticleAlibaba Cloud Alibaba GTE / gte-multilingual-base - nolist.ainolist.ai · undated
ArticleGenerating 50k+ code embeddings in 25 seconds using GTE and MapReducedocs.sweep.dev · 2023-08-09 · 4 HN pts
ArticleGenerating 50k+ embeddings in 25 seconds using GTE and MapReducedocs.sweep.dev · 2023-08-09 · 1 HN pts

Show all 7

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: 17 quantized builds and 61 Spaces.

As a service0no API provider listed
On own hardware17quantized builds
Built on108107 fine-tunes · 1 adapters · 0 merges
In apps61Hugging Face Spaces

Top Spaces running it: antoinelouis/mtem-pruner ♥ 14 · MathematicalModelingAgent/MathematicalModelingAgent ♥ 9 · denniscraandijk/mtem-pruner ♥ 3 · jaaaacky/MathematicalModelingAgent ♥ 1

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, 0.7 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.

#23 of 64 on MTEB so far

62.51. Each tick is a model on this site with a measured MTEB row; author-reported numbers stay in Reference. Mean over 39/41 tasks present. 39/41 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 (2024-07-20), amber triangles are 4 dated articles (each links to it), grey ticks are repo commits.

What people built on it

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

125 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 keisuke-miyako/gte-multilingual-reranker-base-gguf for llama.cpp, Ollama, LM Studio 613 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	50.87 nDCG@10 x100 (mean of 10 tasks)	10 tasks, split=test	leaderboard	2026-10-24
measured MTEB · Embed text for search	62.51 mean over 39/41 tasks present	39/41 tasks	leaderboard	2026-10-24
measured MTEB Reranking · Rerank search results	46.81 MAP x100 (mean of 2 tasks)	2 tasks, split=test	leaderboard	2026-10-24

What the card says

2 cited sentences

Summary generated, cited

Alibaba-NLP/gte-multilingual-base is a 305M parameter encoder-only transformer model designed for sentence similarity tasks.^[1]^[2]^[3] The model supports over 70 languages and accommodates text lengths up to 8192 tokens.^[4]^[5]

parameters human
pipeline tag
“encoder-only transformers architecture”
“Supports over 70 languages”
“up to 8192 tokens”

Generated 2026-10-07 15:02 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

21 fields

Listed in Embeddings

Facts

gte-multilingual-base has 305M parameters. (safetensors metadata, observed 2026-10-24)
Context length is 8,192 tokens (8K). (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 2024-07-20, last modified 2025-07-05. (HF API, observed 2026-10-24)

Pipeline tag	sentence-similarity
Library	sentence-transformers
Architectures	NewModel, NewForTokenClassification
Model type	new
Dtype	float16
Params by dtype	F16: 305,369,089
Size band	S
MoE	no
Context length	8,192 (config.json max_position_embeddings)
Languages	af, ar, az, be, bg, bn, ca, ceb, cs, cy, da, de, el, en, es, et, eu, fa, fi, fr, gl, gu, he, hi, hr, ht, hu, hy, id, is, it, ja, jv, ka, kk, km, kn, ko, ky, lo, lt, lv, mk, ml, mn, mr, ms, my, ne, nl, no, pa, pl, pt, qu, ro, ru, si, sk, sl, so, sq, sr, sv, sw, ta, te, th, tl, tr, uk, ur, vi, yo, zh
Papers	arXiv:2407.19669, arXiv:2210.09984, arXiv:2402.03216, arXiv:2007.15207, arXiv:2104.08663, arXiv:2402.07440
License	Apache 2.0 (apache-2.0)
Hub snapshot	2026-10-25

Servability

vLLM · SGLang

Servability

vLLM: not in registry · 7050ef881ff0835f6890a47db1859a24c65481d6SGLang: not in registry · 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 gte-multilingual-base: params × bytes/param × 1.2 overhead. KV cache and activations are extra. How this is computed.

BF16 (2 bytes/param)	0.7 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: Alibaba-NLP/gte 125 derivatives on the Hub

quantized: 17finetune: 107adapter: 1

Top variants (40 shown, ranked by downloads)

Variant	Relation	Quant	Params	Band	License	Author	Likes	Downloads
onnx-community/gte-multilingual-base	quantized	—	—	—	—	onnx-community	13	323.5k
Trendyol/TY-ecomm-embed-multilingual-base-v1.2.0	finetune	—	305M	S	apache-2.0	Trendyol	42	40.2k
xmanii/maux-gte-persian-v2	finetune	—	305M	S	—	xmanii	6	37.5k
SteveTran/ob_semantic_model	finetune	—	305M	S	—	SteveTran	0	34.2k
xmanii/maux-gte-persian-v3	finetune	—	305M	S	—	xmanii	6	24.6k
upskyy/gte-base-korean	finetune	—	305M	S	apache-2.0	upskyy	4	21.9k
xmanii/maux-gte-persian	finetune	—	305M	S	—	xmanii	8	18.3k
newmindai/TurkEmbed4STS	finetune	—	305M	S	—	newmindai	2	17.1k
louisbrulenaudet/lemone-embed-pro	quantized	—	305M	S	apache-2.0	louisbrulenaudet	2	9.9k
Dqdung205/medical_vietnamese_embedding	finetune	—	305M	S	—	Dqdung205	1	4k
bi-matrix/G-MATRIX-Embedding-v1	finetune	—	305M	S	—	bi-matrix	0	2.3k
xmanii/maux-gte-persian-v3-fp16	finetune	-fp16	305M	S	—	xmanii	1	1.4k
antoinelouis/french-mgte-base	quantized	—	142M	S	apache-2.0	antoinelouis	1	942
BookingCare/gte-multilingual-base-v2.1	finetune	—	305M	S	—	BookingCare	0	964
BookingCare/gte-multilingual-base-v1	finetune	—	305M	S	—	BookingCare	1	898
jaeyong2/gte-multilingual-base-Thai-embedding	finetune	—	305M	S	apache-2.0	jaeyong2	0	926
Sampath1987/EnergyEmbed-v1	finetune	—	305M	S	—	Sampath1987	0	909
impresso-project/OCR-robust-gte-multilingual-base	finetune	—	305M	S	agpl-3.0	impresso-project	1	834
mjwong/gte-multilingual-base-xnli	finetune	—	306M	S	apache-2.0	mjwong	0	858
skfrost19/reranker-gte-multilingual-base-msmarco-bce	finetune	—	306M	S	—	skfrost19	0	768
mjwong/gte-multilingual-base-xnli-anli	finetune	—	306M	S	cc-by-nc-4.0	mjwong	1	708
impresso-project/histlux-gte-multilingual-base	finetune	—	305M	S	agpl-3.0	impresso-project	0	711
denniscraandijk/dutch-gte-multilingual-base	quantized	—	143M	S	apache-2.0	denniscraandijk	0	702
keisuke-miyako/gte-multilingual-reranker-base-gguf	quantized	gguf	—	—	—	keisuke-miyako	1	613
juyoungml/gte-base-ko	finetune	—	305M	S	—	juyoungml	2	539
impresso-project/OCR-diversely-robust-gte-multilingual-base	finetune	—	305M	S	agpl-3.0	impresso-project	1	450
tboquet/m2v-gte-multilingual-base-pca-128	finetune	—	32.0M	S	mit	tboquet	0	492
impresso-project/histlux_ocr_error_denoising_lrec	finetune	—	305M	S	agpl-3.0	impresso-project	1	442
ferrisS/german-english-gte-multilingual-base	quantized	—	150M	S	apache-2.0	ferrisS	0	446
qxuanson/setfit-gte-base-20-shot	finetune	—	305M	S	—	qxuanson	0	445
anhtuansh/gte-multilingual-legal-1e	finetune	—	305M	S	—	anhtuansh	1	328
jaeyong2/gte-multilingual-base-Ko-embedding	finetune	—	305M	S	apache-2.0	jaeyong2	1	326
jaeyong2/gte-multilingual-base-Ja-embedding	finetune	—	305M	S	apache-2.0	jaeyong2	0	371
RamzyBakir/arabic-gte-multilingual-base-100k	finetune	—	305M	S	apache-2.0	RamzyBakir	0	365
RamzyBakir/arabic-gte-multilingual-base-200k-lora	finetune	—	305M	S	apache-2.0	RamzyBakir	0	348
AhmedZaky1/DIMI-embedding-v2	finetune	—	305M	S	—	AhmedZaky1	0	339
minhdang/gte-base-law-matryoshka	finetune	—	305M	S	—	minhdang	1	266
Sampath1987/EnergyEmbed-1E	finetune	—	305M	S	—	Sampath1987	0	300
jaeyong2/gte-multilingual-base-Id-embedding	finetune	—	305M	S	apache-2.0	jaeyong2	0	294
tboquet/gte-multilingual-base-pca-256	finetune	—	64.0M	S	mit	tboquet	1	227

Repository changelog

28 commits

Repository changelog 28 commits

Date	Commit	SHA
2025-07-05	Add Text Embeddings Inference (TEI) tag & snippet (#28)	9bbca17
2025-03-17	Update README.md (#26)	9fdd4ee
2025-01-09	Update README.md	ca1791e
2025-01-09	Update README.md	85f13cd
2024-11-19	Update README.md (#16)	ade1467
2024-11-16	Usage infinity (#15)	3f01372
2024-09-17	Fix broken SentenceTransformer snippet; format code with Python format (#11)	7fc0678
2024-08-17	Update README.md	f7d567e
2024-08-09	Update README.md	139a658
2024-08-09	Update README.md	eed89c8
2024-08-09	Update README.md	d8367eb
2024-08-09	Upload 5 files	7bcec33
2024-08-09	Update README.md	0880701
2024-08-09	Upload 5 files	9e1001d
2024-08-09	Delete images	cf8484f
2024-08-09	Update README.md	a3d553d
2024-08-09	Update README.md	167d84d
2024-08-09	Upload 5 files	087a024
2024-08-06	Update README.md	2098722
2024-07-30	Update README.md	0822b78
2024-07-30	Update README.md	a6e575d
2024-07-27	Update scripts/gte_embedding.py	af8c4b6
2024-07-27	Update README.md	2d7b768
2024-07-27	Create scripts/gte_embedding.py	0119b51
2024-07-25	Update README.md	7ee2ea3
2024-07-25	Upload config.json	586352f
2024-07-25	[add file]	11922d3
2024-07-20	initial commit	f9741dd

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

GTE-Multilingual Series: A Key Model for Retrieval- ...

alibabacloud.comofficialabout the gte family2,706 words

Alibaba's Tongyi Lab released the GTE-Multilingual (mGTE) series to improve retrieval-augmented generation with multilingual support and long-context handling. These encoder-only models handle up to 8,000 tokens and support 75 languages while offering adjustable vector dimensionality from 128 to 768. The weakly supervised pre-training phase utilized 2.8 billion multilingual text pairs collected from the web to build basic embedding capabilities.written, reviewed, cited

A 128-dimensional embedding saves 6 times more storage than a 768-dimensional embedding with less than 2% loss in retrieval performance.
The training data is truncated to 2,000 tokens first and then extended to 8,000 tokens to increase efficiency during pre-training.

Analysis7

Gte-go – my over-optimized embedding model

rcarmo.github.io2026-04-27about the gte familyHN 4 pts · 1 comments

Rcarmo/gte-go: Golang inference for the GTE Small embedding model

github.com2026-04-25authorityabout the gte familyHN 1 pts · 0 comments597 words

The default build produces a fully self-contained static binary with no C dependencies and no gonum in the hot path. Batching reduces jitter 3–5×, achieving a standard deviation of 2.4ms on amd64 and 4.7ms on arm64.generated, cited

gte-multilingual-base by Alibaba-NLP: Open-Weight Model

savrn.com2,592 words

Alibaba-NLP released gte-multilingual-base, an open-weight model for sentence similarity under the Apache 2.0 license. It features 305M parameters with an 8,192-token context and requires about 0.7 GB of GPU memory at 16-bit precision. The model draws 1.1M downloads a month and supports over 70 languages.written, reviewed, cited

The model uses an encoder-only architecture instead of decode-only LLM architecture to lower hardware requirements.
Inference speed increases by 10x compared to previous decode-only LLM architecture models like gte-qwen2-1.5b-instruct.
The model supports elastic output dense representation while maintaining effectiveness to significantly reduce storage costs.

gte-multilingual-base by Alibaba-NLP | Mixpeek Model Hub

mixpeek.com598 words

Alibaba-NLP released gte-multilingual-base, an embedding model supporting over 70 languages with an 8192-token context window. The model contains 305,369,089 parameters and outputs 768-dimensional vectors, requiring trust_remote_code to load due to its custom architecture. It employs CLS pooling rather than mean pooling to ensure retrieval accuracy on long multilingual passages.written, reviewed, cited

Alibaba Cloud Alibaba GTE / gte-multilingual-base - nolist.ai

nolist.ai

Generating 50k+ code embeddings in 25 seconds using GTE and MapReduce

docs.sweep.dev2023-08-09about the gte familyHN 4 pts · 0 comments

Generating 50k+ embeddings in 25 seconds using GTE and MapReduce

docs.sweep.dev2023-08-09about the gte familyHN 1 pts · 0 comments

Papers

6 papers

Papers 6

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).

mGTE: Generalized Long-Context Text Representation and Reranking Models for Multilingual Text Retrieval this model

Xin Zhang, Yanzhao Zhang, Dingkun Long +10 · 2024-07-29 · PDF · HF paper page (26 upvotes)
Making a MIRACL: Multilingual Information Retrieval Across a Continuum of Languages this model

Xinyu Zhang, Nandan Thakur, Odunayo Ogundepo +6 · 2022-10-18 · PDF · HF paper page (2 upvotes)
MKQA: A Linguistically Diverse Benchmark for Multilingual Open Domain Question Answering this model

Shayne Longpre, Yi Lu, Joachim Daiber · 2020-07-30 · PDF · HF paper page
M3-Embedding: Multi-Linguality, Multi-Functionality, Multi-Granularity Text Embeddings Through Self-Knowledge Distillation cited in card

Jianlv Chen, Shitao Xiao, Peitian Zhang +3 · 2024-02-05 · PDF · HF paper page (10 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)
Benchmarking and Building Long-Context Retrieval Models with LoCo and M2-BERT cited in card

Jon Saad-Falcon, Daniel Y. Fu, Simran Arora +2 · 2024-02-12 · PDF · HF paper page (1 upvotes)

Similar models

3 models

Similar models

embeddinggemma-300m · 303M
same size band (S), same task (sentence-similarity), more providers
paraphrase-multilingual-mpnet-base-v2 · 278M
same size band (S), same task (sentence-similarity), more providers, permissive license
multilingual-e5-base · 278M
same size band (S), same task (sentence-similarity), more providers, permissive license

Sources

3 sources

Sources

Hub snapshot: 2026-10-25 · HF API: 2026-10-24 21:35 UTC · Providers: 2026-10-24 21:44 UTC · Articles: 2026-10-24 21:45 UTC · Summary: 2026-10-07 15:02 UTC · Last verified 2026-10-25

Follow gte-multilingual-base

Watch this model

/feeds/models/Alibaba-NLP--gte-multilingual-base.xml(RSS 2.0) · every change to gte-multilingual-base
/feeds/families/Alibaba-NLP--gte.xml(RSS 2.0) · every change across the gte family

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