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 verify it, not a single ‘gold passage.’

Is it deployable? Yes, as a self-hosted preview. Weights are on Hugging Face under the MIT license. Loading requires transformers>=5.4.0 with trust_remote_code=True. It is not yet available on the Perplexity API. The model card notes that weights and interface may change without backward compatibility.

Why the gold passage falls short

RAG systems split long documents into chunks. A chunk often depends on an entity, heading, or definition stated elsewhere. Contextual models address this with late chunking. The document is encoded in one pass, then pooled per chunk.

Training, however, usually marks one gold chunk per query. Every other chunk becomes a negative, including the sentences that make the answer checkable. Perplexity lists three more problems. Binary labels give a coarse signal. LLM annotation cost grows linearly with dataset size. Labels are also tied to one chunking strategy.

How the training works

The teacher is Perplexity’s query-aware context compression. Training treats retrieval as evidence selection: the model learns to return the answer and the supporting context together, rather than a single isolated passage.

This approach aligns with broader 2026 trends in context engineering, where retrieval systems are expected to provide verifiable evidence for generated answers. The MIT-licensed release and Hugging Face availability signal a continued push toward open, self-hostable RAG infrastructure.

For developers, the key takeaway is that pplx-embed-v2-context-9b-preview changes what an embedding model optimizes for: not just finding the best passage, but finding the answer and the evidence that supports it.