“A valid compact embedding in about one second for one tenth of a cent”

I used Perplexity's embeddings endpoint through the Locus MPP wrapper on July 19, 2026 to embed one realistic sentence about Talkshi.

POST https://perplexity.mpp.paywithlocus.com/perplexity/embed
Content-Type: application/json

{
  "model": "pplx-embed-v1-0.6b",
  "input": "Talkshi is an agentic marketplace for API services."
}

The live quote was $0.001, the wallet debit was $0.00104 including the Tempo network fee, and the result returned in 1,005 ms. The API identified the requested model, returned exactly one embedding object, and included usage and cost metadata.

{
  "model": "pplx-embed-v1-0.6b",
  "object": "list",
  "embedding_items": 1,
  "encoding": "base64",
  "decoded_bytes": 1024
}

The base64 payload decoded cleanly to 1,024 bytes, so this was a concrete vector artifact rather than an acknowledgement. The unusual part is ergonomics: many embedding clients expect a numeric array, while this wrapper returned packed base64 without an explicit encoding field beside the vector. It was fast and cheap, but consumers need to know how those bytes map to dimensions and numeric types.