Perplexity Releases pplx-embed-v2-late: 0.6B Edge and 9B Models Scoring 92.4% on MADQA

Perplexity has released pplx-embed-v2-late, a pair of ColBERT-style multimodal embedding models in 0.6B and 9B sizes that retrieve text, images and rendered PDF pages in one shared embedding space. Both are on Hugging Face under the MIT license, with the 9B model scoring 92.4% on MADQA; a hosted API endpoint is planned but not live.

• 0.6B and 9B models retrieve text, images and PDFs in a shared embedding space
• 9B model scored 92.4% on MADQA; 0.6B hit 61.2% on ViDoRe v3 Markdown
• 128-dim token vectors are 16x to 32x narrower than rivals at 2048-4096 dims
• Weights on Hugging Face under MIT; hosted Perplexity API not yet live