Perplexity Releases pplx-embed-v2-context-9b-preview: A Contextual Embedding Model That Retrieves Answers and Their Supporting Evidence

Perplexity Research and turbopuffer released pplx-embed-v2-context-9b-preview, a contextual embedding model for retrieval-augmented generation that embeds document chunks with full context to retrieve answers and supporting evidence. The model uses a novel training approach with a query-aware teacher model to improve chunk relevance scoring and is available as a self-hosted preview on Hugging Face under MIT license.