RAG and data · 2026-09-22
Calculating the cost of indexing millions of chunks with Voyage 4 Lite
Covers that voyageai/voyage-4-lite runs at one sixth of Voyage 4 Large's price, how that price difference changes total cost in a large-scale indexing job, and when that savings makes sense.
The unit price is small, but the gap grows with scale
voyageai/voyage-4-lite is priced as low as $0.02 per million tokens; the catalog description frames this by stating it runs at one sixth of Voyage 4 Large's price. For vectorizing a single document, that difference is negligible, but in a job that needs to process millions of chunks — log search, product catalog matching, or large-scale document indexing — that sixth-of-the-price ratio directly shows up in the total bill.
The 32,000-token context window and flexible output dimensions from 256 to 2048 are the same as Voyage 4 Large's; so there's no trade-off on those two technical specs — the difference is only in the semantic quality of the vector the model produces (the catalog description doesn't state this as a direct number, you need to test it with your own task).
Calculate the economy of scale with your own workload
When indexing a collection of millions of chunks, estimating total token count and comparing Voyage 4 Lite's price against Voyage 4 Large's shows the absolute size of the savings. For example, in a 10-billion-token indexing job, cost with Voyage 4 Lite comes to around $200, while a model at six times the unit price could push that to $1,200 — the difference grows in absolute terms as scale grows.
This calculation isn't just for the initial indexing; it's a recurring cost line in a system where the collection is regularly updated (new documents added, old ones reprocessed). In a system doing regular re-indexing, the unit-price difference compounds its effect over time.
- Vectorizing a single document: the price difference is negligible.
- Batch indexing millions of chunks: the sixth-of-the-price ratio directly saves on the total bill.
- In systems doing regular re-indexing, the difference compounds over time.
Verify the quality ceiling with your own search results
A lower price can mean some trade-off in search quality; before moving to a large-scale job, comparing Voyage 4 Lite's and a higher-tier Voyage model's hit rate on a representative sample is more reliable than deciding on price alone. In a forgiving usage pattern like log search, a small quality difference may not matter, while on a precision-sensitive product-matching task the same difference can affect user experience.
Frequently asked questions
Is Voyage 4 Lite's output-dimension flexibility the same as Voyage 4 Large's?
Yes, both support 256, 512, 1024, and 2048 dimensions; there's no difference on this technical spec.
Is Voyage 4 Lite advantageous at every scale?
The price difference translates into a meaningful absolute saving only at high volume (millions of chunks); on a small collection the difference can stay negligible — decide based on your quality needs.