Skip to main content

Shared Embedding Space

All Voyage 4 series models (voyage-4-large, voyage-4, voyage-4-lite, voyage-4-nano) share the same embedding space. This means vectors from different models are directly comparable — you can embed documents with one model and queries with another.

Why This Matters​

The shared embedding space enables asymmetric retrieval — a cost optimization strategy:

  • Embed your document corpus once with a cheaper model (e.g., voyage-4-lite at $0.02/1M tokens)
  • Embed queries with a higher-quality model (e.g., voyage-4-large at $0.12/1M tokens)

Since queries are short and infrequent compared to your entire corpus, this dramatically reduces costs while maintaining search quality.

Cost Savings​

# See the savings for your workload
vai estimate --docs 1M --queries 10M --months 12

# The numbers:
# Symmetric (voyage-4-large for everything): ~$180/year
# Asymmetric (lite for docs, large for queries): ~$38/year
# That's ~79% savings

How It Works​

Voyage 4 models are trained to place text in the same semantic region regardless of which model in the family generates the embedding. A document embedded with voyage-4-lite occupies nearly the same position as it would with voyage-4-large.

The practical implication: cosine similarity between a voyage-4-large query vector and a voyage-4-lite document vector is meaningful and produces good retrieval results.

Try It​

# Benchmark cross-model similarity
vai benchmark asymmetric

# Compare costs
vai estimate --doc-model voyage-4-lite --query-model voyage-4-large

# Learn more
vai explain shared-space

Further Reading​