OpenRouter adds six Voyage AI embedding, reranking models
OpenRouter introduced six new Voyage AI models covering text embeddings, multimodal embeddings, and instruction-following rerankers. The release features 32K context windows, shared vector spaces across embedding models, and pricing from $0.02 per million tokens.
Bringing specialized embedding and reranking models into OpenRouter significantly simplifies building cost-effective, multi-stage RAG pipelines under a single API framework.
- –Shared vector spaces across Voyage 4 models enable hybrid indexing strategies (e.g., cheap indexing with lite, high-precision retrieval with large) without re-indexing datasets.
- –Instruction-following rerankers allow dynamic natural-language relevance scoring instead of static cosine similarity.
- –Multimodal 3.5 bridges text and rich visual document elements like tables, figures, and slides into a single embedding space.
- –Highly aggressive pricing ($0.02-$0.12/M tokens) makes production-scale enterprise search and retrieval far more economical.
DISCOVERED
1h ago
2026-07-29
PUBLISHED
1h ago
2026-07-29
RELEVANCE
AUTHOR
OpenRouter