Skip to main content

Embedding & Reranking

Embedding Models

Model NameModel IDDimensionsMax Context
BGE-m3bge-m31024 (fixed)8K tokens
Qwen3-VL-Embedding 8Bqwen3-vl-embedding-8b128 – 4096 (configurable)32K tokens

Qwen3-VL-Embedding 8B

Multimodal embedding model supporting text, image, and mixed inputs. Compatible with Cohere's query/document differentiation on the embed endpoint for improved retrieval.

Supported dimensions: 128, 256, 384, 512, 768, 1024, 1536, 2048, 2560, 3072, 3584 and 4096.


Reranking Model

Model NameModel IDDescription
BGE-Reranker-v2-m3bge-reranker-v2-m3Classifies and re-scores embeddings from a vector database on each RAG query