"MongoDB is taking a new step toward AI developers with the general availability of the Voyage 4 embedding family and expansion of its startup program. The integration with Voyage AI, acquired last year, should make it easier for developers to take applications from prototype to production. MongoDB is increasingly positioning itself as the foundation for AI stacks rather than just a database."
"The embedding models are now available via APIs in the managed MongoDB Atlas service and in the on-premises community edition. Embeddings are numerical representations of data that capture semantic meaning as vectors. This allows systems to compare and retrieve information based on meaning rather than exact keywords. The Voyage 4 series consists of four variants, each with its own balance between accuracy, latency, and cost."
"MongoDB says the models improve accuracy for production AI workloads. This is because data no longer needs to be moved or duplicated between separate systems. For developers, this means less complexity and faster implementation. In addition, the company announced the general availability of voyage-multimodal-3.5. This model extends support for text and images to video. "This unlocks unified retrieval across multiple content types," says Franklin Sun, staff product manager. "You have one embedding model instead of three to handle different data types.""
MongoDB released the Voyage 4 embedding family and expanded its startup program while integrating Voyage AI to streamline moving applications from prototype to production. The embedding models are available via APIs in MongoDB Atlas and on-premises community edition. Embeddings represent data as vectors that capture semantic meaning, enabling retrieval by meaning instead of exact keywords. The Voyage 4 series includes standard, large, lite, and nano variants balancing accuracy, latency, cost, and local development. The company also made voyage-multimodal-3.5 generally available, extending embeddings to support text, images, and video for unified retrieval across content types. The models reduce data movement, lowering complexity and accelerating implementations.
Read at Techzine Global
Unable to calculate read time
Collection
[
|
...
]