Voyage AI

Voyage AI provides embedding and reranking models accessible through a simple API, tuned specifically for retrieval quality rather than general-purpose representation. Its models consistently rank among the top performers on retrieval benchmarks, which is why many teams reach for Voyage embeddings when retrieval accuracy is the bottleneck in their RAG system. Voyage AI was acquired by MongoDB in 2025.

It is aimed at engineers building search and RAG who want strong out-of-the-box retrieval without training their own models. Beyond general-purpose embeddings, Voyage ships domain-specific variants (code, finance, law) and rerankers that reorder candidate passages by true relevance.

Highlights

• Retrieval-tuned embeddings — general-purpose and domain-specific models that punch above their size on retrieval tasks.
• Rerankers — cross-encoder models that take a query plus candidate passages and return them sorted by relevance.
• Long context & flexible dimensions — large input lengths and configurable output dimensions to trade quality against storage cost.
• Quantization-friendly — int8 and binary output options to shrink vector storage in your database.

In an AI-assisted workflow

Embed documents at ingestion and the query at search time, store the vectors in a database like Qdrant, then optionally rerank the candidates:

import voyageai
vo = voyageai.Client()  # reads VOYAGE_API_KEY
 
doc_vectors = vo.embed(chunks, model="voyage-3", input_type="document").embeddings
query_vector = vo.embed([question], model="voyage-3", input_type="query").embeddings[0]
# ...nearest-neighbour search in your vector DB, then:
reranked = vo.rerank(question, candidates, model="rerank-2", top_k=5)

Good to know

Voyage AI is a hosted API with a free tier of monthly tokens to start, then usage-based pricing. Because embeddings from different models are not interchangeable, switching embedding models later means re-embedding (and re-indexing) your whole corpus — see Choosing Embeddings in 2026 before you commit.