Perplexity releases pplx-embed-v2-late multimodal embedding models
This is a developer-facing embedding model release for teams building their own search or retrieval-augmented systems that need to match text and images together — it has no direct impact on how someone uses the Perplexity app itself, and there is no date yet for when it reaches Perplexity's own API.
What changed
Source summary
Perplexity released pplx-embed-v2-late, a late-interaction embedding model family in two sizes (0.6B and 9B parameters): both checkpoints share the same multi-vector embedding space, so the smaller model's queries can search an index built by the larger model. The models separately support semantic image search with text queries, including searching PDF pages without OCR. Both models are publicly available on Hugging Face now for developers to use in their own search or retrieval-augmented systems; Perplexity says it will progressively roll out support for these embeddings on its own API Platform, with no date given.
Why it matters to your work
Practical interpretation
This is a developer-facing embedding model release for teams building their own search or retrieval-augmented systems that need to match text and images together — it has no direct impact on how someone uses the Perplexity app itself, and there is no date yet for when it reaches Perplexity's own API.
When this could help
Relevant when choosing or evaluating an embedding model for a custom retrieval pipeline that needs to search across both text and images.
pplx-embed-v2-late provides text and image retrieval in 0.6B and 9B model sizes for developers building their own search systems. It can search rendered PDF pages without requiring OCR or parsed text.
How to get started
Both the 0.6B and 9B pplx-embed-v2-late models are available on Hugging Face. The announcement names the smaller model pplx-embed-v2-late-0.6b.
Provider guidance ↗For the documented Python integration, use sentence-transformers version 6.0.0 or later and transformers version 5.4.0 or later; the announcement states the models are compatible out of the box with these versions.
Provider guidance ↗
What to check
- This is a developer model release on Hugging Face. The announcement describes Perplexity API Platform support as a future rollout, not an available API feature. Integrating the models requires your own retrieval system. Source ↗
Sources & provider guides
Official documentation checked 13 Oct 2026. SCALA prompts and workflows are suggestions, not claims of hands-on testing.
Source & availability
This is a summary of a published provider announcement, with practical context from SCALA’s feed.
Plan, region, workspace and rollout conditions may affect access in Australia. Check the original source for current availability.
www.perplexity.aiThis update is primarily for technical teams.