Perplexity Open Sources 9B and 0.6B Embedding Models for OCR-Free Search

Perplexity has open-sourced two late-interaction embedding models, pplx-embed-v2-late in 9B and 0.6B parameter versions, that enable OCR-free search across text and images in a shared vector space. The models, hosted on Hugging Face and supported in Sentence Transformers, scored 92.4% on MADQA and 81.3 nDCG@10 on text retrieval tasks for the 9B version. Perplexity reports that pairing the 0.6B query model with 9B indexing improves text retrieval to 79.6 nDCG@10.