Skip to main content

Choosing a Model

Use this decision guide to select the right Voyage AI model for your use case.

Quick Decision Tree​

General-Purpose: Voyage 4 Family​

NeedModelWhy
Best retrieval qualityvoyage-4-largeHighest RTEB score (71.41), MoE architecture
Balance quality + costvoyage-4Strong quality (70.07) at half the price
Minimize costvoyage-4-liteGood quality (68.10) at $0.02/1M
Asymmetric retrievalvoyage-4-large (query) + voyage-4-lite (docs)Best quality queries, cheapest storage

Domain-Specific Models​

Use these when your content is primarily in one domain — they outperform general-purpose models on domain-specific benchmarks:

  • voyage-code-3 — Code, technical documentation, API references
  • voyage-finance-2 — Financial reports, SEC filings, earnings calls
  • voyage-law-2 — Legal documents, contracts, case law

Reranking Models​

NeedModel
Best reranking qualityrerank-2.5
Fast reranking / lower costrerank-2.5-lite

Cost Comparison​

# Compare costs for your specific workload
vai estimate --docs 500K --queries 2M --months 12

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

Tips​

  • Start with voyage-4 (balanced) and benchmark against voyage-4-large using vai eval — upgrade only if the quality difference matters for your use case.
  • Always use reranking in production (rerank-2.5). The cost is minimal and the quality improvement is significant.
  • Domain-specific models are worth the premium only if your content is primarily in that domain. For mixed content, use general-purpose models.

Further Reading​