bge-base-en

BAAI MIT out 3 years ago

109M parameters, text in, embeddings out, 512 tokens of context. 1 provider serves it.

Verified 2026-10-25 Rank #51 in Embeddings · score 88.9 method v1 Sources 3 Downloads 30d 2.5M · #56 of 619 tracked

Can I run it?0.1GB at INT4Estimatefits a 16 GB cardmethod

Can I use it?MITPermissivelicence file2026-10-24

Does it work?72.33on MTEB#4 of 64 measuredleaderboard2026-10-24

Do people use it?2.5Mdownloads, 30 days#56 of 619 we track · top 9%+2 likes, downloads +3% this week72 likesHugging Face2026-10-24

Worth reading

Picked by rule from 6 tracked pieces: the lab's own post, the most shared, a field report, the latest.

Latest · 6 months ago

BGE Embeddings Guide: Complete Tutorial | AI Wiki

BGE embeddings are open-source text models from BAAI designed for retrieval and RAG pipelines.

artificial-intelligence-wiki.com 2026-05-04 2.5k words about the bge family family

6 best of 6 tracked

The 1 pick above, plus these 5. Top is the radar's score: discussion, source authority and the lab's own posts, fading with age.

ArticleBGE Explanation — BGE documentationbge-model.com · undated
ArticleBGE family (BAAI) | Semantic Similarity Intermediate Course | The Neural Basetheneuralbase.com · undated
ArticleFine-tuning — BGE documentationbge-model.com · undated
Articlebge-base-en-v1.5 by BAAI, Feature Extraction | Mixpeek Model Hubmixpeek.com · 2026-04-25
DiscussionShow HN: Mixture of Voices–Open source goal-based AI router-uses BGE transformernews.ycombinator.com · 2025-09-17 · 1 HN pts

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	HF Inference	—	—	—	—	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; 3 quantized builds for local use.

As a service1HF Inference
On own hardware3quantized builds
Built on4141 fine-tunes · 0 adapters · 0 merges
In apps66Hugging Face Spaces

Top Spaces running it: embedl/hfviewer ♥ 181 · nickmuchi/Earnings-Call-Analysis-Whisperer ♥ 39 · zakaria-ch49/Medical-Question-Answering-with-Retrieval-Augmented-Generation_RAG ♥ 2 · kizabgd123/ai-workflow-orchestrator ♥ 2

Will it fit?

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

Fits a 16 GB card at INT4

About 0.1 GB at INT4, 0.1 GB at FP8, 0.3 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.

#4 of 64 on MTEB so far

72.33. Each tick is a model on this site with a measured MTEB row; author-reported numbers stay in Reference. Mean over 27/41 tasks present. 27/41 tasks

Source · leaderboard · 2026-10-24

The story so far

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

3 dated articles and 21 repo commits so far

Teal is the weights (2023-08-05), amber triangles are 3 dated articles (each links to it), grey ticks are repo commits.

What people built on it

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

44 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 mradermacher/bge-base-en-i1-GGUF for llama.cpp, Ollama, LM Studio 2.3k downloads

All variants →

Reference

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

Benchmarks

2 rows

Benchmarks 2 rows

measured third-party leaderboard row (2)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	72.33 mean over 27/41 tasks present	27/41 tasks	leaderboard	2026-10-24
measured MTEB Reranking · Rerank search results	46.49 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 model maps text to low-dimensional dense vectors for tasks such as retrieval, classification, clustering, or semantic search.^[1]

“map any text to a low-dimensional dense vector”

Generated 2026-10-07 15:04 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

bge-base-en has 109M 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 2023-08-05, last modified 2024-04-17. (HF API, observed 2026-10-24)

Pipeline tag	feature-extraction
Library	transformers
Architectures	BertModel
Model type	bert
Dtype	float32
Params by dtype	F32: 109,482,240 · I64: 512
Size band	S
MoE	no
Context length	512 (config.json max_position_embeddings)
Languages	en
Papers	arXiv:2310.07554, arXiv:2309.07597
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 bge-base-en: params × bytes/param × 1.2 overhead. KV cache and activations are extra. How this is computed.

BF16 (2 bytes/param)	0.3 GB · Edge (<4 GB)
FP8 (1 byte/param)	0.1 GB · Edge (<4 GB)
INT4 (0.5 bytes/param)	0.1 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: BAAI/bge 44 derivatives on the Hub

finetune: 41quantized: 3

Top variants (40 shown, ranked by downloads)

Variant	Relation	Quant	Params	Band	License	Author	Likes	Downloads
Ionio-ai/retail_embedding_classifier_v1	finetune	—	109M	S	—	Ionio-ai	0	49.2k
mradermacher/bge-base-en-i1-GGUF	quantized	GGUF	—	—	mit	mradermacher	0	2.3k
Xenova/bge-base-en	quantized	—	—	—	—	Xenova	0	1.5k
mradermacher/bge-base-en-GGUF	quantized	GGUF	—	—	mit	mradermacher	0	1.4k
riphunter7001x/bge-base-financial	finetune	—	109M	S	apache-2.0	riphunter7001x	1	758
ivanleomk/finetuned-bge-base-en	finetune	—	109M	S	—	ivanleomk	0	465
dshvadskiy/finetuned-bge-base-en	finetune	—	109M	S	—	dshvadskiy	0	417
swathibp/BGE-base_finetuned	finetune	—	109M	S	—	swathibp	0	392
ppuva1/finetuned-bge-base-en	finetune	—	109M	S	—	ppuva1	0	379
paulperry/finetuned-bge-base-en	finetune	—	109M	S	—	paulperry	0	352
apostolosfilippas/finetuned-bge-base-en	finetune	—	109M	S	—	apostolosfilippas	0	352
aaa961/bge-base-en-cost-categories-3sets	finetune	—	109M	S	apache-2.0	aaa961	0	348
dakini/finetuned-bge-base-en	finetune	—	109M	S	—	dakini	0	333
Maximgolubov/finetuned-bge2-base-en	finetune	—	109M	S	—	Maximgolubov	0	328
lzwcv/finetuned-bge-base-en	finetune	—	109M	S	—	lzwcv	0	327
llazio/finetuned-bge-base-en	finetune	—	109M	S	—	llazio	1	272
labdmitriy/finetuned-bge-base-en	finetune	—	109M	S	—	labdmitriy	0	315
RK-1235/bge-base-FIR-matryoshka-BASELINE-10epo-triplet	finetune	—	109M	S	—	RK-1235	0	299
RK-1235/bge-base-FIR-matryoshka-BASELINE-10epochs-FT	finetune	—	109M	S	apache-2.0	RK-1235	0	298
trasul/bge-base-all-nli-triplet	finetune	—	109M	S	—	trasul	0	296
nhegde/finetuned-bge-base-en	finetune	—	109M	S	—	nhegde	0	285
aaa961/finetuned-bge-base-en	finetune	—	109M	S	—	aaa961	0	285
jkdamilola/finetuned-bge-base-en	finetune	—	109M	S	—	jkdamilola	0	284
Oussama1209/model_mnlp_encoder	finetune	—	109M	S	—	Oussama1209	0	279
TahmidTapadar/finetuned-bge-base-en	finetune	—	109M	S	—	TahmidTapadar	0	277
3lv27/finetuned-bge-base-en	finetune	—	109M	S	—	3lv27	0	275
alexisbrooker/finetuned-bge-base-en	finetune	—	109M	S	—	alexisbrooker	0	252
aaa961/finetuned-bge-base-en-firefox-bugs-bugs	finetune	—	109M	S	—	aaa961	0	249
juanpprim/finetuned-bge-base-en	finetune	—	109M	S	—	juanpprim	0	202
bishaln/finetuned-bge-base-en	finetune	—	109M	S	—	bishaln	0	186
trasul/mpnet-base-all-nli-triplet	finetune	—	109M	S	—	trasul	0	163
aaa961/finetuned-bge-base-en-firefox	finetune	—	109M	S	—	aaa961	0	142
aaa961/finetuned-bge-base-en-firefox-bugs-pairs_10	finetune	—	109M	S	—	aaa961	0	138
curiousabhinav/finetuned-bge-base-en	finetune	—	109M	S	—	curiousabhinav	0	137
RK-1235/bge-base-FIR-matryoshka-BASELINE-10epochs-finetune	finetune	—	109M	S	apache-2.0	RK-1235	0	125
RK-1235/bge-base-FIR-matryoshka	finetune	—	109M	S	apache-2.0	RK-1235	0	124
vijay-delete/finetuned-transaction-classification-bge-base-en	finetune	—	109M	S	—	vijay-delete	0	124
aaa961/finetuned-bge-base-en-firefox-bugs-pairs	finetune	—	109M	S	—	aaa961	0	111
RK-1235/bge-base-financial-matryoshka	finetune	—	109M	S	apache-2.0	RK-1235	0	103
RishuD7/bge-base-en-41-keys-phase-2-v1	finetune	—	109M	S	apache-2.0	RishuD7	0	102

Repository changelog

21 commits

Repository changelog 21 commits

Date	Commit	SHA
2024-04-17	Adding ONNX file of this model (#6)	b737bf5
2023-10-12	update	17d124a
2023-09-27	update	5e2f5ae
2023-09-15	update	6747e44
2023-09-12	update	03bc1d2
2023-09-12	update	436526f
2023-09-12	update	f858c91
2023-09-12	update	75d0064
2023-09-12	update	0725d0e
2023-09-11	Merge branch 'main' of https://huggingface.co/BAAI/bge-base-en into main	a6d78b9
2023-08-19	add FAQ	05c14ca
2023-08-15	Adding `safetensors` variant of this model (#4)	90e113f
2023-08-09	update readme	0cceaac
2023-08-09	update readme	a268908
2023-08-09	update readme	2f68ed7
2023-08-06	Update README.md	1241160
2023-08-05	Update README.md	922064b
2023-08-05	Update README.md	787f6bc
2023-08-05	Update config.json	8375bab
2023-08-05	upload	d17f9a1
2023-08-05	initial commit	1d87ec1

All providers

1 provider · 1 endpoint

Where to run 1 providers · 1 endpoints

Hosted endpoints observed for bge-base-en, sorted by input price (USD per million tokens). Each row names its observer and date; prices change daily, so check the source before quoting.

Provider	Model ref	Context	Quant	$/M in	$/M out	tok/s	TTFT	Tools	Status	Observed
HF Inference hf	BAAI/bge-base-en	—	—	—	—	—	—	—	live	2026-10-25

Articles and coverage

6 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.

Analysis5

BGE Embeddings Guide: Complete Tutorial | AI Wiki

artificial-intelligence-wiki.com2026-05-04about the bge family2,538 words

BGE embeddings are open-source text models from BAAI designed for retrieval and RAG pipelines. The BGE-M3 model supports up to 8192 tokens and handles over 100 languages, while BGE-VL is the multimodal variant introduced in March 2025. English-specific variants like bge-large-en-v1.5 offer 1.34 GB models with 1,024 dimensions for high-accuracy tasks.written, reviewed, cited

BGE Explanation — BGE documentation

bge-model.comabout the bge family675 words

BGE-v1.5 utilizes a BERT-base structure featuring 12 encoder layers and a hidden dimension of 768 to generate embeddings. The model is trained to use the last hidden state of the [CLS] token as the sentence embedding, whereas using mean pooling results in a significant decrease in performance. Installation is available through the FlagEmbedding library or via the Hugging Face transformers package using the BAAI/bge-base-en-v1.5 identifier.written, reviewed, cited

The model tokenizer configuration sets a maximum input length of 512 tokens for sentence encoding.
The intermediate linear layer expands the hidden dimension from 768 to 3072 before projection.
The pooling function supports two methods: extracting the [CLS] token or computing the mean of all tokens.

BGE family (BAAI) | Semantic Similarity Intermediate Course | The Neural Base

theneuralbase.comabout the bge family1,227 words

BAAI's BGE family provides Apache 2.0 licensed open-source embeddings that match proprietary APIs at 1/10th the inference cost by utilizing hard negative mining during training. The bgebaseenv1.5 model, containing 334M parameters, remains the recommended production variant for English-only search despite newer multilingual options. This approach shifts semantic search from an API-centric problem to an infrastructure problem, enabling teams to self-host without sacrificing quality.written, reviewed, cited

Model.encode() defaults to CPU inference; add device='cuda' parameter to achieve the claimed 100x cheaper production speed.
bgelargeenv1.5 provides ~23% better retrieval at 1.3x latency, but the improvement may not justify the inference cost increase.
Fine-tuning BGE requires implementing hard negative mining, or you will regress performance even with domain-specific examples.

Fine-tuning — BGE documentation

bge-model.comabout the bge family26,859 words

Engineers can fine-tune BGE models using the FlagEmbedding package with the finetune dependency. The tutorial demonstrates training the BAAI/bge-large-en-v1.5 model with a learning rate of 1e-5 over 2 epochs. Required data must include query, pos, and neg fields, while optional arguments control sequence length and knowledge distillation.written, reviewed, cited

The system automatically sets the deepspeed accelerator to cuda if detected on the hardware.

bge-base-en-v1.5 by BAAI, Feature Extraction | Mixpeek Model Hub

mixpeek.com2026-04-25616 words

The model produces 768-dimensional vectors over a 512-token window with 109.5 million parameters. It averages 63.55 on the MTEB English table, which includes 56 English datasets.generated, cited

Community1

Show HN: Mixture of Voices–Open source goal-based AI router-uses BGE transformer

news.ycombinator.com2025-09-17about the bge familyHN 1 pts · 1 comments726 words

Papers

2 papers

Papers 2

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).

A Multi-Task Embedder For Retrieval Augmented LLMs cited in card

Peitian Zhang, Shitao Xiao, Zheng Liu +2 · 2023-10-11 · PDF · HF paper page (6 upvotes)
C-Pack: Packed Resources For General Chinese Embeddings cited in card

Shitao Xiao, Zheng Liu, Peitian Zhang +3 · 2023-09-14 · PDF · HF paper page (2 upvotes)

Similar models

3 models

Similar models

bge-base-en-v1.5 · 109M
same size band (S), same task (feature-extraction), more providers, permissive license
canine-c · 132M
same size band (S), same task (feature-extraction), permissive license
bart-base · 139M
same size band (S), same task (feature-extraction), permissive license

Sources

3 sources

Sources

Hub snapshot: 2026-10-25 · HF API: 2026-10-24 21:37 UTC · Providers: 2026-10-24 21:44 UTC · Articles: 2026-10-24 21:45 UTC · Summary: 2026-10-07 15:04 UTC · Last verified 2026-10-25

Follow bge-base-en

Watch this model

/feeds/models/BAAI--bge-base-en.xml(RSS 2.0) · every change to bge-base-en
/feeds/families/BAAI--bge.xml(RSS 2.0) · every change across the bge family

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