voyage-4

Voyage-4 is a core member of a new generation of text embedding models engineered to enhance retrieval accuracy for production-ready AI systems. Designed with a shared embedding space, the model allows developers to move away from the traditional constraint of using identical models for both document indexing and query generation. This architectural flexibility enables teams to optimize their infrastructure by pairing different models within the same series to balance retrieval quality, latency, and operational costs according to specific use case requirements.

The model benefits from advanced training techniques, including Matryoshka Representation Learning, which supports flexible embedding dimensions to maintain high performance across various scales. By utilizing quantization-aware training, the model achieves efficient output handling while preserving retrieval precision. These design choices make it a robust choice for developers building context-engineered agents and systems requiring long-term memory, as it allows for seamless interoperability between high-fidelity indexing and high-throughput query tasks.