How billing works
Retrieval exposes two separate primitives: embeddings, measured by tokens, and reranking, measured by calls.
- Embedding price follows the number of tokens processed.
- Reranking has a fixed per-call price.
- Semantic storage search is a separate platform operation.
Price in practice
Embedding work is billed by the volume processed, while reranking is a request-level operation. A retrieval workflow can use one or both, so its total depends on which primitives it invokes.
Embedding and reranking pricing does not replace the 1-credit semantic storage-search operation.
Rate card
ServicePriceHow it is applied
Nemotron 3 Embed 1B10 CrPer 1M embedding tokens
Nemotron 3 Embed 1B reranking4 CrPer call
Questions
Are vector searches themselves charged?
Semantic storage search is a separate platform operation and costs 1 Cr per successful search.
Can I use embeddings without reranking?
Yes. They are distinct operations with separate billing units.
Does embedding pricing include semantic storage search?
No. Semantic storage search is a separate platform operation.