Voyage AI produces the highest-scoring embedding models on the MTEB benchmark as of 2026. The voyage-3-large model leads retrieval accuracy across 55 datasets. Voyage also offers domain-specific variants (voyage-code-3 for code, voyage-finance-2 for finance, voyage-law-2 for legal) that outperform general models on their respective domains.
Drop-in replacement for OpenAI or Cohere embeddings — same API structure (POST text, get back vectors). Supports 32K token inputs (longer than OpenAI). Domain-specific variants are fine-tuned on curated data for code, legal, finance, etc., so they handle domain-specific vocabulary better than general models. Acquired by Anthropic in 2024.
voyage-3-lite: $0.02 per M tokens. voyage-3: $0.06 per M tokens. voyage-3-large: $0.12 per M tokens. voyage-code-3, voyage-finance-2, voyage-law-2: $0.12 per M tokens.
Teams prioritizing retrieval accuracy, domain-specific RAG (legal, financial, code), and research groups. Anthropic uses Voyage for Claude's retrieval.