PerplexitySpecialist modelOpen weights
pplx-embed-v2-context-9b
Summary
Perplexity announced pplx-embed-v2-context with open 0.6B and 9B weights on November 3, 2026. The multilingual vision-language model supports retrieval across text, images, and visual documents, including PDF pages.
Contextual mode embeds text chunks with the rest of the document in view. Text also has a non-contextual mode, and image pages can be split into horizontal chunks. The 9B checkpoint produces int8 vectors with 2,048 dimensions and supports Matryoshka prefixes of 1,024 or 2,048 dimensions.
Its 1,024-dimensional prefix shares an embedding space with the 0.6B variant. The pplx-embed-v2-context-9b-preview checkpoint is a separate earlier preview whose card warns against mixing its vectors with later models. The 9B weights use MIT according to the model card. A hosted API or price is not established.
Similar models
More models of the same kind as pplx-embed-v2-context-9b, chosen by model line, developer, size, price and release date.
- pplx-embed-v2-late-9bPerplexity · same developer
- pplx-embed-v2-context-9b-previewPerplexity · same developer
- pplx-embed-v2-context-0.6bPerplexity · same developer
- pplx-embed-v2-late-0.6bPerplexity · same developer
- AstaBrief-8BAllen Institute for AI · similar size
- FLUX 3 ActionBlack Forest Labs · similar size
- Qwen-Drive-1.0-4BQwen · similar size
- Unlimited OCR 2Baidu · similar size
All specifications
Architecture
Input and output
- Modalities
- Text, images, and visual documents. Supports contextual and non-contextual text inputs as well as contextual image chunks.
- Source: Modalities
Watermarking and provenance
AI watermarking refers to invisible signals embedded in the output. Visible marks and provenance metadata are listed separately. Unknown means there is insufficient evidence for that particular field.
- AI watermarking
- Unknown
- no source
- Watermarking technique
- Not known
- no source
Licence and access
- Shutdown
- None announced
- no source
Further details
- Embedding output
- One 2,048-dimensional int8 vector per document or document chunk. The model card specifies Matryoshka prefixes of 1,024 and 2,048 dimensions.
- Source: Embedding output
- Shared embedding space
- The 1,024-dimensional Matryoshka prefix shares its embedding space with the 0.6B variant.
- Source: Shared embedding space
- Earlier preview
- The existing pplx-embed-v2-context-9b-preview checkpoint is a separate preview version. Its model card warns against mixing its embeddings with later versions.
- Source: Earlier preview