Perplexity’s models search PDFs, slides and scans without OCR

Perplexity says its models embed images and rendered pages directly, allowing PDFs, slides and scans to be searched without OCR. The company says its 9B and 0.6B models share an embedding space after distillation from one 18B teacher. It claims queries using the 0.6B model on a 9B-indexed corpus lift ViDoRe v3 from 62.3% to 63.5% with no added query cost.