voyage-context-4
Voyage Context 4 embeds chunks with their document context. It introduces a mixture-of-experts backbone and automatic chunking.
Model design
The release evaluates both chunk retrieval and document retrieval across 39 datasets in eight domains. Its long-document workflow divides text into smaller parts. This does not imply that a single attention pass has an unlimited context window.voyage-context-4 and automatic chunking
Request limits
The API documents a 32,000-token limit per chunk. All inputs in one request must total no more than 120,000 tokens with automatic chunking, or 32,000 tokens without it. Automatic chunking is off by default; its default chunk size is 512 tokens.Voyage AI contextualized chunk embedding API
Practical use
Practical guidance. Keep the parent document ID and chunk position with each vector. Retrieve the matching passage and nearby text before composing an answer. Automatic splitting saves preparation work, but the returned passage still needs its source title and location.
Specifications & access
| Detail | Value and conditions |
|---|---|
| Launch date | |
| Access | API |
| Atlas lifecycle | Active from 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-4-large
Voyage AI · Voyage 4
Mixture-of-experts text embeddings with a shared Voyage 4 vector space.
Launch date API - voyage-context-3
Voyage AI · Voyage Context
Document-aware chunk embeddings trained for chunk and document relevance.
Launch date API