Skip to content

Voyage 4

Voyage 4 is a mid-sized model in Voyage AI's Voyage 4 family. Voyage AI reports it approaches voyage-3-large retrieval quality with a context window of 32K tokens, Matryoshka dimensions (2048, 1024, 512, 256), and multiple quantization options. All Voyage 4 models share one embedding space, so you can mix models for asymmetric retrieval.

index.ts
import { embed } from 'ai';


const result = await embed({
  model: 'voyage/voyage-4',
  value: 'Sunny day at the beach',
})

What To Consider When Choosing a Provider

  • Configuration: When query volume dominates cost, embed the corpus once with voyage-4-large and serve queries with voyage-4-lite, or Voyage 4. Voyage AI reports higher accuracy than symmetric retrieval with smaller models alone.
  • Configuration: Use Voyage 4 for both queries and documents when you want one model and balanced cost.
  • Configuration: Moving from voyage-3.5, voyage-3-large, or older models requires re-embedding because the embedding space differs from Voyage 4.
  • Zero Data Retention: AI Gateway does not currently support Zero Data Retention for this model. See the documentation for models that support ZDR.
  • Authentication: AI Gateway authenticates requests using an API key or OIDC token. You do not need to manage provider credentials directly.

When to Use Voyage 4

Best For

  • General-purpose retrieval: You want Voyage 4's shared space and mid-sized efficiency
  • Asymmetric setups: Documents use voyage-4-large and queries use Voyage 4 to control latency and cost
  • RAG pipelines: Use Matryoshka dimensions and quantization to cut vector database cost
  • Teams moving from voyage-3-large: Want Voyage 4 compatibility and flagship accuracy on the document side

Consider Alternatives When

  • Highest average retrieval scores in Voyage AI's published Voyage 4 benchmarks: Use voyage-4-large (MoE flagship)
  • Lowest compute for queries: Use voyage-4-lite when per-query cost is the binding constraint and voyage-3.5-level accuracy is sufficient
  • Source-code-only corpora: voyage-code-3 stays purpose-built for code
  • Multimodal text-and-image embeddings: Pick a model with native image inputs

Conclusion

Voyage 4 is the practical default for teams that want Voyage 4 quality without flagship compute costs. The shared embedding space means you can start here and layer in voyage-4-large for documents or voyage-4-lite for high-volume queries without re-indexing.

Frequently Asked Questions

  • What is the difference between Voyage 4, voyage-4-large, and voyage-4-lite?

    voyage-4-large is the MoE flagship with the highest average retrieval scores in Voyage AI's published Voyage 4 benchmarks. Voyage 4 is the mid-sized model; Voyage AI reports it approaches voyage-3-large quality. voyage-4-lite uses fewer parameters; Voyage AI reports it approaches voyage-3.5 retrieval accuracy. All three share one embedding space.

  • How does Voyage 4 compare to voyage-3.5?

    Voyage 4 is a Voyage 4 model with a shared embedding space and updated training. Voyage AI positions voyage-4-lite near voyage-3.5 accuracy; Voyage 4 targets voyage-3-large-level quality. Moving from Voyage 3.x requires re-embedding your corpus.

  • What is the context window for Voyage 4?

    32K tokens. Set chunk sizes so single-pass embeds stay under this limit on long texts.

  • Can I use Voyage 4 for RAG applications?

    Yes. Voyage 4 is a text embedding model for semantic search and retrieval-augmented generation across mixed content types, including technical documentation, business text, and conversational text.

  • How do I access Voyage 4 through Vercel AI Gateway?

    Add your Voyage AI API key in AI Gateway settings, then send embedding requests through AI Gateway. AI Gateway authenticates requests and records usage.

  • Do I need to re-embed my data to switch from Voyage 3.x to Voyage 4?

    Yes. Voyage 3 and Voyage 4 use different embedding spaces, so you re-embed and re-index when you move generations. Within Voyage 4, you can often change query models without re-vectorizing documents if you follow Voyage AI's asymmetric retrieval pattern with voyage-4-large document embeddings.

  • What is shared embedding space in Voyage 4?

    All Voyage 4 models map text into the same vector space, so embeddings from different models in the family are compatible. You can search document vectors from voyage-4-large with query vectors from Voyage 4 or voyage-4-lite.