jina-embeddings-v2-small-en

A compact English-only text embedding model from Jina AI that encodes sentences and documents up to 8,192 tokens into 512-dimensional dense vectors for semantic search, RAG, and similarity matching. Built on a JinaBERT backbone with symmetric bidirectional ALiBi positional encoding, it delivers strong long-context retrieval performance at only 33M parameters. Source: supabase.