Announcing New Models and Expanded Availability

At Voyage AI, we empower developers to build applications that rely on highly accurate AI-powered search and retrieval. We’re committed to building state-of-the-art models that maximize retrieval accuracy while offering unparalleled cost optimization features that serve every developer, no matter where they are building.

Today, we are excited to make a number of announcements:

The Voyage 4 model series: Introduces a shared embedding space, eliminating the need to re-index your data when switching between models in the series. This series also includes voyage-4-nano, our first open-weights model.
voyage-multimodal-3.5: Extends our industry-leading multimodal capabilities to video retrieval.

Expanded distribution: Newly available on MongoDB Atlas and GCP, with additions to existing model availability on AWS and Azure.

The Voyage 4 Series: A New Frontier in Retrieval Accuracy

voyage-4-large supersedes voyage-3-large as our most accurate model, replacing it as the top model on the RTEB leaderboard. voyage-4 and voyage-4-lite offer additional price-performance flexibility, while voyage-4-nano is our first open-weights model available on Hugging Face.

The Voyage 4 series introduces a shared embedding space, allowing cross-model compatibility for all embeddings generated within the Voyage 4 series. This provides unparalleled flexibility to prioritize accuracy while minimizing latency and reducing cost. Use different Voyage 4 models across pipeline stages, such as using voyage-4-large for high-fidelity indexing, voyage-4-lite for high-throughput queries, and voyage-4-nano for local development.

Learn more in the Voyage 4 release blog.

voyage-multimodal-3.5: Introducing Video Retrieval

voyage-multimodal-3.5 builds upon our existing multimodal support for text and content-rich images by adding native video retrieval capabilities, enabling semantic search over video content using natural language queries. Additionally, voyage-multimodal-3.5 is the first production-grade embedding model to support Matryoshka embeddings for flexible dimensionality.

Learn more in the voyage-multimodal-3.5 release blog.

Expanded Availability

Voyage AI models are now accessible across multiple platforms, allowing you to use our embedding models and rerankers with your existing cloud credits and infrastructure:

Stay connected with Voyage AI

Follow us on X (Twitter) and LinkedIn for more updates.