Pricing & billing
One balance, shared with Hindsight Cloud
The Inference API draws on your organization's single credit balance — the same balance that pays for Hindsight Cloud memory operations. There is no separate inference wallet to top up, and no separate invoice. Add credits from Billing in either the Inference console or the Hindsight Cloud console; both spend and refill the same balance.
Rates
Every endpoint bills on input tokens only — output is free.
| Model | Used for | Price (input) |
|---|---|---|
hs-embed-1 | Embeddings | $0.005 / 1M tokens |
hs-decide-1-flash (default) | Reranking & decision | $0.02 / 1M tokens |
hs-rerank-1 (cross-encoder) | Reranking | $0.015 / 1M tokens |
The live rates are also returned per model by GET /v1/models under pricing.input_per_million, and each response reports what it was billed (usage/meta.billed_units).
How a charge works
Before each request, the API checks your balance against the most the request could cost. If the balance is too low it returns 402 without running the model or charging anything. Otherwise the model runs and the actual token cost is deducted from your balance.
When the shared balance runs low or reaches zero because of an inference call, the organization's owners get a low-balance or credits-exhausted email so inference and memory keep working.
Notes
- Decision requests (
POST /v1/systemone) are billed at the reranking rate — they run the same model. - Chat/generation models are internal to Hindsight and are not sold through the API, so there is no per-token generation charge to plan for at launch.