
Perplexity: Embed V2 Context 0.6B
perplexity/pplx-embed-v2-context-0.6b
Model weightspplx-embed-v2-context-0.6b is a compact multimodal contextual embedding model from Perplexity for retrieval over text, images, and visual documents. It is the cost-efficient sibling of pplx-embed-v2-context-9bOpens in new tab and is suited for high-volume RAG and search indexing where chunks of the same document should be embedded with their surrounding context in view: pass a document as an ordered list of chunks and each chunk's embedding reflects the whole document, or embed strings individually for standard dense retrieval.
Queries and documents use separate input_type modes, so send query for search queries and document for the content being indexed. It returns 1024-dimensional vectors that share an embedding space with the 9B model's 1024-dimensional prefix, so documents indexed with one model can be queried with the other. Image input is accepted in document mode only.