voyage-context-3
Voyage Context 3 produces one vector per chunk while using the surrounding document. Training combines document-level and chunk-level relevance signals.
Model design
Unlike a token-level multi-vector index, this design keeps one embedding for each supplied chunk. Related chunks enter the model as one document group. The release supports variable vector lengths and quantized representations.voyage-context-3 contextualized chunk embeddings
Practical use
Practical guidance. Use it when an isolated paragraph loses a definition, heading or reference from elsewhere in the document. Keep document boundaries intact in each request. Joining unrelated sources into one group can give the encoder misleading context.
Specifications & access
| Detail | Value and conditions |
|---|---|
| Launch date | |
| Access | API |
| Atlas lifecycle | Legacy since 30 June 2026MongoDB Voyage AI model lifecycle: Atlas scope and release history |
| Output dimensions | 1,024 default; 256, 512 or 2,048 optionalVoyage AI contextualized chunk embedding API |
Related models
- voyage-context-4
Voyage AI · Voyage Context
Chunk embeddings that use surrounding document context and automatic splitting.
Launch date API - voyage-3-large
Voyage AI · Voyage 3
General embeddings with selectable dimensions and quantized output.
Launch date API