By No Code MBA · Facts checked
- Provider
- Perplexity
- Announcement date
- Status at last check
- Limited access or preview
- Model family
- pplx-embed
- Preceding model
- pplx-embed-context-v1
- Capabilities
- Text, Embeddings
What changed
pplx-embed-v2-context-9b-preview is a roughly 9B contextual embedding model that encodes document chunks with document-wide context and returns one embedding per chunk. It outputs 2048-dimensional vectors, with trained 1024-dimension truncation and native int8, and was trained on about 430 datasets in over 50 languages. Perplexity reports 45.5% answer recall at K=10 on turbopuffer's private context-bench, 14.4 points above voyage-context-4, and the top average nDCG@10 on ConTEB among models shown.
Access and availability
Preview weights are on Hugging Face at perplexity-ai/pplx-embed-v2-context-9b-preview under the MIT license, run locally with transformers 5.4.0 or later. Perplexity says API access is in progress.
Official access documentationWhat you can build
Test it on long documents where answer passages depend on distant context, comparing retrieval of both answer-bearing and supporting passages against your current chunk embeddings.
Limitations
Perplexity warns weights, embeddings and interface may change without backward compatibility, so preview embeddings should not be mixed with later versions. The main benchmark is private and run by a co-author, and the model trails voyage-context-4 slightly on average for query-to-document retrieval.