GTE
Alibaba Cloud'/%3e%3cdefs%3e%3cpattern%20id='pattern0_1240_293949'%20patternUnits='userSpaceOnUse'%20patternTransform='matrix(11.2%200%200%2011.2%203.93617%200)'%20preserveAspectRatio='none'%20viewBox='0%200%2016%2016'%20width='1'%20height='1'%3e%3cg%20id='pattern0_1240_293949_inner'%3e%3crect%20width='16'%20height='16'%20fill='%23f88376'%20mask='url(%23hatch_gaps)'/%3e%3cmask%20id='hatch_gaps'%20maskUnits='userSpaceOnUse'%20x='0'%20y='0'%20width='16'%20height='16'%3e%3crect%20width='16'%20height='16'%20fill='white'/%3e%3crect%20x='6'%20width='2'%20height='2'%20fill='black'/%3e%3crect%20x='6'%20y='8'%20width='2'%20height='2'%20fill='black'/%3e%3crect%20x='4'%20y='2'%20width='2'%20height='2'%20fill='black'/%3e%3crect%20x='4'%20y='10'%20width='2'%20height='2'%20fill='black'/%3e%3crect%20x='2'%20y='4'%20width='2'%20height='2'%20fill='black'/%3e%3crect%20x='2'%20y='12'%20width='2'%20height='2'%20fill='black'/%3e%3crect%20y='6'%20width='2'%20height='2'%20fill='black'/%3e%3crect%20y='14'%20width='2'%20height='2'%20fill='black'/%3e%3crect%20x='14'%20width='2'%20height='2'%20fill='black'/%3e%3crect%20x='14'%20y='8'%20width='2'%20height='2'%20fill='black'/%3e%3crect%20x='12'%20y='2'%20width='2'%20height='2'%20fill='black'/%3e%3crect%20x='12'%20y='10'%20width='2'%20height='2'%20fill='black'/%3e%3crect%20x='10'%20y='4'%20width='2'%20height='2'%20fill='black'/%3e%3crect%20x='10'%20y='12'%20width='2'%20height='2'%20fill='black'/%3e%3crect%20x='8'%20y='6'%20width='2'%20height='2'%20fill='black'/%3e%3crect%20x='8'%20y='14'%20width='2'%20height='2'%20fill='black'/%3e%3c/mask%3e%3c/g%3e%3c/pattern%3e%3c/defs%3e%3c/svg%3e)
GTE (General Text Embedding) covers embedding and reranking for multilingual and long-document retrieval. gte-multilingual-base handles more than 70 languages at 8,192 tokens from a 305M-parameter encoder and emits sparse token weights alongside dense vectors; the ModernBERT-based embedder and cross-encoder reranker target English long documents, and the gte-Qwen2 instruct checkpoints are decoder-based variants. Alibaba's NLP group publishes the family.
The gme-Qwen2-VL checkpoints are a separate multimodal line and are not counted in this record.
Openness
3 high confidence- weights
- open(Apache-2.0 ungated checkpoints across the multilingual, ModernBERT and Qwen2-based tiers)
- data
- closed(the cards point at the mGTE paper for the training mixture and release no corpus)
- code
- partial(a gte_embedding.py inference script ships in the model repository
- license
- Apache-2.0(OSI)
The category center of gravity: an OSI license over ungated weights with the corpus and the training run kept in the paper. The Qwen2-derived checkpoints carry Apache-2.0 rather than the Qwen license, so no restrictive SKU pulls the family down.
- https://huggingface.co/Alibaba-NLP/gte-modernbert-base/raw/main/README.md recorded 2026-09-11
The embedding checkpoint's own card - BEIR average 55.33, LoCo long-context and CoIR code-retrieval results at 8,192 tokens, and the statement that the series follows the previous GTE training scheme with ModernBERT as the base, referring readers to the mGTE paper for training details.
- https://huggingface.co/Alibaba-NLP/gte-multilingual-base/raw/main/README.md recorded 2026-09-11
Describes the training architecture and points to the mGTE paper for detail, releases no corpus, and ships scripts/gte_embedding.py as inference code for dense and sparse output.
- https://huggingface.co/api/models/Alibaba-NLP/gte-modernbert-base recorded 2026-09-11
license: apache-2.0; gated: false; private: false.
- https://huggingface.co/api/models/Alibaba-NLP/gte-multilingual-base recorded 2026-09-11
license: apache-2.0; gated: false; private: false; downloadable checkpoint files present.
- https://huggingface.co/api/models/Alibaba-NLP/gte-Qwen2-7B-instruct recorded 2026-09-11
license: apache-2.0; gated: false; private: false - the Qwen2-derived tier is distributed under Apache-2.0 rather than the Qwen license.
Adoption
4 high confidenceAdoption is measured as Hugging Face downloads across the GTE embedding checkpoints in the Alibaba-NLP account, led by gte-multilingual-base. The two reranker checkpoints are scored separately as gte-reranker and are excluded here, as are the older v1 encoders published under the separate thenlper account.
- https://huggingface.co/api/models/Alibaba-NLP/gte-modernbert-base recorded 2026-09-11
downloads: 171930; likes: 200; license apache-2.0 - one of the embedding checkpoints the summed figure counts.
- https://huggingface.co/api/models/Alibaba-NLP/gte-multilingual-base recorded 2026-09-11
downloads: 1307627.
- https://huggingface.co/api/models/Alibaba-NLP/gte-Qwen2-7B-instruct recorded 2026-09-11
downloads: 111229.
Capability
4 high confidenceGTE has the highest BEIR average among sub-1B models in its own published table, with long context throughout and a multilingual encoder alongside the English one. It falls short of the leader, which is measured on MMTEB rather than BEIR.
- https://huggingface.co/Alibaba-NLP/gte-modernbert-base/raw/main/README.md recorded 2026-09-11
The embedding checkpoint's own card gives gte-modernbert-base MTEB-en 64.38 and BEIR 55.33 at 8,192 maximum sequence length, with LoCo long-context and CoIR code-retrieval results on the same table.
- https://huggingface.co/Alibaba-NLP/gte-multilingual-base/raw/main/README.md recorded 2026-09-11
States support for over 70 languages, 8192-token input, elastic dense output and sparse vectors, with retrieval validated on MIRACL, MLDR, MKQA, BEIR and LoCo.
Verified 2026-09-11