voyage-code-4
Voyage Code 4 targets code retrieval from natural-language requests. Its training data includes completed pull requests that link issue descriptions to changed code.
Model design
The release evaluates 19 new agent-oriented retrieval tasks alongside 28 earlier code datasets. The new tasks use held-out repositories and files changed in issue-fixing pull requests. Training covers hundreds of programming languages, but this is not a guarantee of equal quality in each language.voyage-code-4: code retrieval for agents
Practical use
Practical guidance. Use issue text or a bug report to retrieve candidate files, then inspect the code before changing it. Keep exact symbol search available for known identifiers. Measure whether the required files appear in the candidate set, not whether a generated fix sounds plausible.
Specifications & access
| Detail | Value and conditions |
|---|---|
| Launch date | |
| Access | API |
| Context window | 32,000 tokens Per input text. MongoDB Voyage AI model specifications |
| Atlas lifecycle | ActiveMongoDB Voyage AI model lifecycle: Atlas scope and release history |
| Output dimensions | 1,024 default; 256, 512 or 2,048 optionalMongoDB Voyage AI model specifications |
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-code-3
Voyage AI · Voyage Code
Code embeddings with variable dimensions and quantization support.
Launch date API - voyage-code-2
Voyage AI · Voyage Code
Earlier code retrieval encoder for natural-language and code queries.
Launch date API