Perplexity Releases pplx-embed-v2-context-9b-preview: A Contextual Embedding Model That Retrieves Answers and Their Supporting Evidence
Perplexity's new embedding model learns to retrieve answers AND their evidence—a shift in how RAG systems are trained.
Why it matters
Perplexity Research released pplx-embed-v2-context-9b-preview, a contextual embedding model that changes RAG training signals: instead of retrieving a single 'gold passage,' it learns to surface both the answer and supporting context needed to verify it. The model embeds each chunk with full document visibility, improving retrieval grounding for production RAG pipelines.
The key facts
14 to knowModel: pplx-embed-v2-context-9b-preview (9B parameters)
Status: Preview (not GA)
Training mechanism: contextual embeddings with full-document awareness, not single-passage gold labels
Use case: RAG pipeline retrieval with evidence grounding
Deployment ready: yes (per article)
Partner: turbopuffer (vector database/infrastructure)
Training signal innovation: dual retrieval (answer + supporting context) vs. traditional single passage
Model: pplx-embed-v2-context-9b-preview (9B parameters, contextual embeddings)
Training signal shift: retrieves answer + supporting context, not single 'gold passage'
Each chunk embedded with full document in view
Status: preview, deployable
Collaboration: Perplexity Research + Turbopuffer
Domain: RAG (retrieval-augmented generation) pipelines
Motivation: grounding answers with verifiable evidence
Go to the source
MarkTechPostmarktechpost.com
Publisher excerpt: Perplexity Research and turbopuffer have released pplx-embed-v2-context-9b-preview, a contextual embedding model for RAG pipelines. Each chunk is embedded with the full document in view. The real change is the training signal. The model learns to retrieve the answer along with the context needed to…