voyage-context-4: stop worrying about chunking with our best-performing model

Voyage-context-4 is an advanced contextualized chunk embedding model that enhances document processing by capturing full document context without manual metadata or context augmentation. It introduces a new mixture-of-experts (MoE) backbone that improves context-aware embeddings, incorporates auto-chunking for documents longer than 32K tokens, and supports overlapping chunks, thereby eliminating the need for a separate chunking strategy. The model outperforms its predecessor, voyage-context-3, by 1.4% on document-level and 2.08% on chunk-level retrieval and surpasses voyage-4-large by 0.4% in single-embedding evaluation. Voyage-context-4 is designed to improve retrieval accuracy across various domains, including technical documentation, medical, and legal fields, and supports different embedding dimensions for customizable quality-storage tradeoffs. As a drop-in replacement for voyage-context-3, it is priced at $0.12 per 1M tokens, offering a cost-effective solution for embedding needs without compromising on performance.

No tracked trend matches for this post yet.