PulseAugur
EN
LIVE 09:58:23

Perplexity releases new embedding models, achieves SOTA on benchmarks

By PulseAugur Editorial · [8 sources] ·

Perplexity has released two new late-interaction embedding models, pplx-embed-v2-late, which can retrieve text, images, and pages within a shared embedding space. These models achieve frontier performance and are available on Hugging Face. The 9B model demonstrated strong results on agentic search tasks, outperforming a ColBERT model on BrowseComp+ questions, and achieved high Recall@1000 scores on the Q2D-Web benchmark. Additionally, Perplexity highlighted that these models can process PDFs, slides, and scans without OCR, preserving layout and tables. AI

IMPACT These new embedding models could enhance search capabilities and cross-modal querying, potentially setting new benchmarks for retrieval systems.

RANK_REASON Perplexity is a frontier lab releasing new embedding models.

Read on X — Perplexity →

AI-generated summary · Google Gemini · from 8 sources. How we write summaries →

Perplexity releases new embedding models, achieves SOTA on benchmarks

How we ranked this

Signal score
0 / 100
Composite score across the factors below. Higher = stronger signal that this story matters right now.
Newsworthiness bucket
Frontier Release
Perplexity is a frontier lab releasing new embedding models.
Source corroboration
8 independent sources
Strong cross-source corroboration — multiple independent publishers covered this within the clustering window.
Topics
model release, product
Editorial topic classification. Feeds into how the story surfaces on /topic/<slug> hub pages and into the per-entity coverage mix.
AI-industry relevance
High
Clearly on-topic for AI-industry coverage.
Story freshness
8 days old
Aged out of breaking-news scoring windows; ranking reflects the durable signal from the full source set.

Full methodology in our editorial standards.

COVERAGE [8]

  1. X — Perplexity TIER_1 English(EN) · perplexity_ai ·

    Both models are on Hugging Face: https://t.co/JHy1DhFaCT

    Both models are on Hugging Face: https://t.co/JHy1DhFaCT

  2. X — Perplexity TIER_1 English(EN) · perplexity_ai ·

    On MADQA, 500 questions over 800 PDFs, it reaches 92.4%, the top result among retrievers. https://t.co/uxW8LDau2c

    On MADQA, 500 questions over 800 PDFs, it reaches 92.4%, the top result among retrievers. https://t.co/uxW8LDau2c

  3. X — Perplexity TIER_1 English(EN) · perplexity_ai ·

    In agentic search, a GPT-OSS-120B agent using the 9B model answers 64.0% of BrowseComp+ questions correctly, 4.9 points ahead of the next ColBERT model, with fe

    In agentic search, a GPT-OSS-120B agent using the 9B model answers 64.0% of BrowseComp+ questions correctly, 4.9 points ahead of the next ColBERT model, with fewer searches than any baseline. https://t.co/hFngH5ES7D

  4. X — Perplexity TIER_1 English(EN) · perplexity_ai ·

    On Q2D-Web, about 70,000 production-derived search queries over 190M web documents, the 9B and 0.6B models reach 74.8% and 73.6% Recall@1000, against 69.3% for

    On Q2D-Web, about 70,000 production-derived search queries over 190M web documents, the 9B and 0.6B models reach 74.8% and 73.6% Recall@1000, against 69.3% for nemotron-embed-8b. https://t.co/FwltQP6jGN

  5. X — Perplexity TIER_1 English(EN) · perplexity_ai ·

    Both sizes are distilled token by token from one 18B teacher, so they share one embedding space.

    Both sizes are distilled token by token from one 18B teacher, so they share one embedding space. A corpus indexed with the 9B model can be searched with 0.6B queries. That lifts ViDoRe v3 from 62.3% to 63.5% with no added query cost. https://t.co/SglIFldtGQ

  6. X — Perplexity TIER_1 English(EN) · perplexity_ai ·

    The models also embed images and rendered pages directly, so PDFs, slides, and scans are searched without OCR, keeping tables, figures, and layout that text ext

    The models also embed images and rendered pages directly, so PDFs, slides, and scans are searched without OCR, keeping tables, figures, and layout that text extraction drops. https://t.co/iZmlOKG1ye

  7. X — Perplexity TIER_1 English(EN) · perplexity_ai ·

    Dense embedding models compress each document into one vector, which loses detail as pages get longer or more visual.

    Dense embedding models compress each document into one vector, which loses detail as pages get longer or more visual. pplx-embed-v2-late keeps a 128-dimensional vector per token and scores with MaxSim, so each query token is matched to its closest token in the document. https://…

  8. X — Perplexity TIER_1 English(EN) · perplexity_ai ·

    We're releasing pplx-embed-v2-late, two late-interaction embedding models that retrieve text, images, and pages with a shared embedding space for cross-model qu

    We're releasing pplx-embed-v2-late, two late-interaction embedding models that retrieve text, images, and pages with a shared embedding space for cross-model querying. Both models achieve frontier performance and are publicly available on Hugging Face. https://t.co/AM7YEyQkKE h…

RELATED ENTITIES

RELATED TOPICS