Voyage 4 shares one embedding space across large, lite and nano

Voyage AI's Voyage 4 family allows different models to handle document indexing and incoming queries. Their numerical representations are compatible: a query encoded by voyage-4-lite can search documents encoded by voyage-4-large. The release explanation calls this a shared embedding space.

That compatibility matters because documents may be indexed infrequently while searches arrive continuously. A team can use the larger model to represent its document collection, then choose a smaller query model to reduce recurring computation. It can later change the query model within the family without recalculating all the document representations, separating two costs that would otherwise be coupled.