2026-10-07 # Perplexity unveils two embedding models, one scoring 92.4% on MADQA Perplexity says its new 9B embedding model scored 92.4% on the MADQA benchmark, alongside a 0.6B edge model, per MarkTechPost. ## What else the source says - Model line named pplx-embed-v2-late - Edge model has 0.6B parameters - Larger model has 9B parameters - 9B model scored 92.4% on MADQA benchmark ## Why it matters Benchmark gains may boost Perplexity's search-ranking ambitions. Source. [news.google.com](https://news.google.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?oc=5) ## TRACE - First seen Oct 8, 2026, 3:30 AM PT - Published Oct 9, 2026, 2:46 AM PT, 23.3 h after first seen - Source news.google.com (Press) [SOTO home](https://signalsoftomorrow.com/)