Pricing & models
Set your max. We find a provider that meets it.
Publishers set their own per-token rates. Consumers can attach a bid (x-max-input-price / x-max-output-price) and we route to a provider whose ask meets it. Published rates per 1,000,000 tokens, charged against a USDC balance you control.
Every model has one published ask here; attach x-max-input-price /
x-max-output-price to refuse anything priced above your cap.
Models
—
Upstreams
—
Settled (24h)
—
Rate card
ask is the published rate. last is the observed 24h settlement for that model,
or — when it has no settled traffic yet.
| Model | Family | Ask in $/Mtok | Ask out $/Mtok | Last 24h |
|---|---|---|---|---|
| Loading the rate card… | ||||
Generated from GET /api/v1/pricing.
Callable model IDs
The exact strings to pass as model. IDs are case-sensitive.
| Model ID | Input $/Mtok | Output $/Mtok |
|---|---|---|
| Loading… | ||
Billing notes
Rates are per 1,000,000 tokens and are the same for input and output on a given model.
Each model carries a weight multiplier; the charge for a request is its token count times that weight.
Deposits are credited at 1 USDC = 1,000,000 units, so a weight of 1 corresponds to $1 per million tokens.
A request that names a model with no published rate is refused with
402 PRICE_UNKNOWN when a cap applies.
Minimum charge: $0.00001 per request. Prices are per million tokens, but a billed request never costs less than $0.00001 — if its tokens come to less (a short prompt on a cheap model, say), it is billed $0.00001. Failed requests and requests with no output cost $0.