Embedding

Last verified

An embedding is a numeric vector representing text, used for semantic search and retrieval.

Embedding models are much cheaper per token than generation models, which makes the per-call cost feel negligible — right up to the point you embed a large corpus. The bill lands in the initial indexing run and again on every re-index after a model change.

Estimate before you run: corpus tokens × rate. And consider the Batch API, since indexing is rarely latency-sensitive.