Models & usage
Which wallet pays for which model, and published rates.
Which wallet pays?
- Caedral-hosted models (caedral-embed-e1-small-v1, caedral-voice-1) → weekly plan credits first, prepaid balance as fallback
- Caedral Rerank (BAAI/bge-reranker-v2-m3) → prepaid balance, $0.002 per request
- All third-party catalog model IDs → prepaid balance only, at provider cost + 5% infrastructure + the capped Notre Fee on eligible chat
- The branded caedral-voice product (third-party gpt-audio) → prepaid balance
- Free-tagged chat models (:free) cost $0 and are fair-use capped per day — never paid usage in disguise
Published Caedral rates
| Product | Model id | Rate |
|---|---|---|
| Caedral Embed | caedral-embed-e1-small-v1 | $0.0025 / 1M tokens |
| Caedral Rerank | BAAI/bge-reranker-v2-m3 | $0.002 / request |
| Caedral Voice V1 | caedral-voice-1 | $0.03 / 1k chars |
| WebSearch | websearch | $0.01 / provider credit used |
Semantic cache savings
For non-streaming chat requests, Caedral keeps a per-account semantic cache of recent completions. A near-exact repeat of a recent question (cosine similarity above 0.97 within 7 days, same model, same account) is served from cache and billed at half the fresh-completion price, with real token counts still logged. Add the header X-Caedral-Cache: off to skip both lookup and storage for a request.