GTE Large is built on a transformer++ encoder backbone that combines BERT architecture with Rotary Position Embedding and Gated Linear Unit components, enabling strong performance across long-document understanding tasks. The model produces 1024-dimensional embeddings optimized for semantic similarity, retrieval, clustering, and ranking scenarios. Developed by Alibaba's Institute for Intelligent Computing, it targets applications requiring nuanced text understanding within a mid-range model size category.

On the MTEB benchmark suite, GTE Large achieves state-of-the-art scores within its size class, with particularly strong results on long-context retrieval tests through the LoCo evaluation. The model is available as open weights under Apache 2.0, integrates with the sentence-transformers ecosystem, and supports deployment via transformers.js. Benchmark results show robust performance on classification tasks including Amazon product reviews, making it practical for production retrieval pipelines, semantic search systems, and document similarity workflows.