pplx-embed-v2-late
pplx-embed-v2-late is Perplexity's family of late-interaction (multi-vector) embedding models for text and images, published October 7, 2026 in 0.6B and 9B sizes.
Provider
Perplexity
Model family
Perplexity Embed
Multimodal embedding model
Cost tier
Embed Late
Status
Current
Release Oct 7, 2026
Technical specs
- Inputs
- text, image
- Outputs
- embedding
- License
- Open weights on Hugging Face; check the model card license
- Sizes
- 0.6B, 9B
- Retrieval
- late interaction (multi-vector)
- Capabilities
- Late-interaction multi-vector embeddings
- Shared text and image embedding space
- 0.6B and 9B sizes
Why teams choose it
Complex reasoning
Useful for workflows that require structured thinking, multi-step logic, and deeper analysis than lightweight models provide.
Embedding throughput and refresh cadence
Useful once you baseline latency, dimensional requirements, corpus refresh schedules, and how often vectors must reconcile with canonical documents.
Perplexity integration posture
Re-read provider limits whenever billing regions or SKU names change—embedding fleets fail quietly when quotas or payload caps shift.
Cost-efficient routing
Useful as part of a routing stack where cheap models handle drafts and confirmations and this tier handles genuinely hard passages.
Tradeoffs to know
- Multi-vector indexes need more storage and a late-interaction retriever, not a plain single-vector store.
- Confirm the license on each Hugging Face model card before commercial use.
Benchmarks
No independent or vendor benchmark scores in this catalog entry.
Related models
Perplexity
Perplexity Decider v1.2
pplx-decider-v1-2 is the 27B model behind Perplexity's Decisions API, released in October 2026 (alias pplx-decider-la…
Perplexity
Perplexity Decider Flash v1.2
pplx-decider-flash-v1-2 is the 4B fast model for Perplexity's Decisions API, released in October 2026 (alias pplx-dec…
pplx-embed-v2-late FAQ
What is pplx-embed-v2-late?
pplx-embed-v2-late is Perplexity's family of late-interaction (multi-vector) embedding models for text and images, published October 7, 2026 in 0.6B and 9B sizes. Text and images share one embedding space, so a text query can retrieve images and the reverse. Perplexity published the weights on Hugging Face.
When does pplx-embed-v2-late fit best?
Multimodal RAG over documents with images
What should teams watch out for with pplx-embed-v2-late?
Multi-vector indexes need more storage and a late-interaction retriever, not a plain single-vector store.
Explore next
Models, tools, and comparisons that connect to this reference.