← Back

Wire · technology

Perplexity Open-Sources pplx-embed-v2-late, 92.4% on MADQA

Published

8 October 2026

Topic

technology

◆ Sectors

AI InfrastructureNLP & Speech

◆ Geography

United States

◆ Source

Read at aiweekly.co →

◆ Verified

Fusion42 · 10 October 2026 · Fusion42 review

Perplexity has open-sourced two multimodal embedding models, a 0.6B parameter edge model and a 9B parameter model sharing one embedding space that enables querying large indexes without rebuilding. The models embed PDFs, slides, and images natively without OCR, scoring 92.4% on MADQA, and support efficient querying for resource-constrained devices, reducing infrastructure costs.

This Wire brief sits within Fusion42's coverage of AI Infrastructure and NLP & Speech.

◆ ◆ The Wire takeaway

You can now serve large, high-quality multimodal document search indexes from a small edge device with no rebuilds, cutting your cloud costs and enabling user-facing queries on constrained hardware. This opens a new phase for retrieval tooling where deployment size and speed no longer force trade-offs with index quality.

◆ Coverage

1 source · 8 Oct 2026

◆ Related on Wire

◆ Topics

AI InfrastructureNLP & Speechopen-sourceembedding-modelsmultimodalretrievaledge-computingcost-reduction