Voyage AI

Embedding model provider. Offers high-dimensional text embedding APIs used in semantic and vector search pipelines.

Referenced in Erik Hatcher’s Hybrid Search Blueprint Series Semantic Boosting as the embedding provider for query vectorization (voyage-4-large, 2048 dimensions, dot product similarity).

Models

• voyage-3-large — used for pre-embedding documents in the embedded_movies example

• voyage-4-large — used for query embedding at query time

• rerank-3-lite — rerankers launched 2026-09-30, succeeding rerank-2.5 / rerank-2.5-lite at the same price ($0.05/$0.02 per M tokens), with 32K context and instruction-following. Largest gains are on long documents and code.

The rerankers were built “in collaboration with the Voyage team at MongoDB” and are also served natively in MongoDB Atlas Vector Search.

People

• Erik Hatcher — used Voyage AI in Semantic Boosting demo

Articles

• Hybrid Search Blueprint Series Semantic Boosting
• Hev meets Jev — rerank-3 scores 0.504 mean nDCG@10 on three BEIR subsets, at $0.50 per 1,000 queries with p50/p95 of 185 ms / 287 ms. That is second to Mixedbread’s mxbai-rerank-v3.1-listwise (0.511) and a hair ahead of an untuned general decision model (Jev, 0.501), with a much tighter latency tail than Jev
• rerank-3 and rerank-3-lite - the next generation of Voyage rerankers — launch post; +2.72% NDCG@10 over Cohere Rerank v4.0 Pro across 95 datasets (vendor-run), and ahead of untuned Jev in code, law and finance

Key Concepts

• Dense Vector Retrieval

• Semantic Boosting

• Embeddings

• Reranking