text-embedding-3-large vs Voyage 4 Large
Which one to choose
Pick OpenAI's text-embedding-3-large when you want a widely used embedding model with shortenable vectors that serves search, clustering, classification and anomaly detection from one API. Pick Voyage 4 Large when retrieval is the job and you want to index documents with the large model but embed live queries with the smaller Voyage 4 or Voyage 4 Lite, since they share one vector space. Neither can mix with the other's vectors.
Pricing comparison
| text-embedding-3-large | Voyage 4 Large | |
|---|---|---|
| Model maker | OpenAI | Other |
| Delivery availability | Available | Available |
| Context window | 8.2K | 32K |
| Max output | - | - |
| Official price | Input$0.13per 1M tokensOutput$0.00per 1M tokens | Input$0.12per 1M tokensOutput$0.00per 1M tokens |
| TokenLab price | Input$0.091per 1M tokensOutput$0.00per 1M tokens | — |
| Model performance | Collecting data | Collecting data |
| Capabilities |
Choose text-embedding-3-large when
- You use embeddings for clustering, classification or anomaly detection as well as search
- You already store OpenAI-style vectors and want a quality upgrade
- You want to trim vector size to fit index limits
Choose Voyage 4 Large when
- Your main workload is multilingual retrieval for RAG or search
- You want cheap query-time embedding against a high-quality document index
- You want several output sizes to trade quality against storage
How they differ
| Aspect | text-embedding-3-large | Voyage 4 Large |
|---|---|---|
| Design focus | OpenAI describes it as its most capable embedding model for English and non-English text. | Voyage describes it as the top retrieval model of its fourth series for general and multilingual text. |
| Mixing model sizes | One model for both documents and queries. | Documents and queries can come from different family members, because the vectors are compatible. |
| Vector size | Dimensions can be shortened to save storage. | Four output sizes are offered, from small to large. |
| Task range | Search, clustering, recommendations, anomaly detection and classification. | Aimed at retrieval and ranking in search and RAG. |
| Switching cost | Vectors are not comparable with other embedding models. | Same rule: moving to or from another family means re-embedding the corpus. |
Summary
- text-embedding-3-large: Input $0.091 / Output $0.00 per 1M tokens; Voyage 4 Large: Input $0.12 / Output $0.00 per 1M tokens. Rates depend on the billing unit, specification, and usage. Compare matching conditions in the model's detailed pricing; a single rate does not determine the total cost.
- Voyage 4 Large has a 3.9x larger context window
FAQ
Which is better for RAG, text-embedding-3-large or Voyage 4 Large?
Voyage 4 Large is the retrieval-focused choice, and its shared space with Voyage 4 Lite lets you cut query cost. text-embedding-3-large is a solid general pick if you also cluster or classify text. Test both on your own queries.
Can I replace one with the other without re-indexing?
No. Each model's vectors live in their own space, so changing models means re-embedding every document and every query before you can compare any results.
Does either handle non-English text?
Yes. OpenAI describes text-embedding-3-large as strong for non-English tasks, and Voyage describes Voyage 4 Large as a multilingual retriever, so both suit corpora that mix several languages.
Can I use both together?
Only in separate indexes. Keep one vector store per model and blend results at the ranking step, never by mixing raw vectors from the two models in one store.
Which is cheaper, text-embedding-3-large or Voyage 4 Large?
text-embedding-3-large: Input $0.091 / Output $0.00 per 1M tokens; Voyage 4 Large: Input $0.12 / Output $0.00 per 1M tokens. Rates depend on the billing unit, specification, and usage. Compare matching conditions in the model's detailed pricing; a single rate does not determine the total cost.
What are the key differences between text-embedding-3-large and Voyage 4 Large?
Both models share similar capabilities.
Sources
Reviewed Oct 2, 2026