卓越
目录平台模板

NVIDIA's generalist embedding model achieving record 69.32 score on MTEB benchmark. Fine-tuned from Llama architecture with improved techniques for training LLMs as embedding models.

State-of-the-art foundational model for ranking with 4096 context length and multilingual support for 100+ languages. Offers exceptional performance on BEIR benchmarks and specialized domains including finance, e-commerce, and enterprise search.

A versatile embedding model from BAAI that simultaneously supports dense retrieval, sparse retrieval, and multi-vector retrieval, with multilingual support for 100+ languages and multi-granularity processing from short sentences to 8192-token documents.

Advanced multi-vector retrieval model creating token-level embeddings with late interaction mechanism, featuring denoised supervision and improved memory efficiency over original ColBERT.

State-of-the-art general-purpose and multilingual embedding model from Voyage AI that ranks first across eight domains spanning 100 datasets, outperforming OpenAI and Cohere models by significant margins.

State-of-the-art multimodal embedding model from BAAI supporting text-to-image, image-to-text, and compositional visual search. Trained on the MegaPairs dataset with over 26 million retrieval triplets.

Multilingual embedding model supporting over 100 languages and ranking #1 on MTEB multilingual leaderboard. Offers flexible model sizes from 0.6B to 8B parameters with user-defined instructions.

Universal multimodal embedding model from Jina AI supporting text and images through unified pathway. Built on Qwen2.5-VL-3B-Instruct, outperforms proprietary models on visually rich document retrieval. This is a commercial API with free tier, though OSS weights available.

First fully reproducible open-source text embedding model with 8,192 context length. v2 introduces Mixture-of-Experts architecture for multilingual embeddings. Outperforms OpenAI models on benchmarks. This is an OSS model under Apache 2.0 license.

A compact and efficient pre-trained sentence embedding model, widely used for generating vector representations of text. It's a popular choice for applications requiring fast and accurate semantic search, often integrated with vector databases.

RaDeR (Reasoning-aware Dense Retrieval) is a research model specifically trained on datasets that require reasoning, enabling it to learn how to retrieve relevant theorems and principles during intermediate reasoning steps. This approach allows the retriever to better generalize to diverse reasoning-intensive retrieval tasks.

High-throughput and memory-efficient open-source LLM inference engine with PagedAttention, continuous batching, and support for embedding model serving. Widely adopted for production-scale AI inference.