Inference usage is deducted from your available credits at the published rates
below. Check your balance and add credits with a card in the
Token Factory dashboard.
Per-million-token rates
The public model catalog is the programmatic source for current rates. Each
model lists
input_rate_cents_per_mtok, cached_input_rate_cents_per_mtok, and
output_rate_cents_per_mtok, and the same prices per token in pricing:
Cached input tokens
When a request starts with a prompt prefix that the model recently processed, the serving engine reuses that work instead of computing it again. Those input tokens are cached input tokens, billed at the cached input rate. Caching is automatic: there is no request parameter, and you do not needcache_control.
Requests that share a long, unchanged prefix, such as a system prompt or earlier
conversation turns, get the most cached input tokens.
Cached input pricing applies to requests that start at or after
2026-10-27 12:00 UTC. Earlier requests were billed at the input rate for all input
tokens.
Every response reports the cached share of its input tokens:
In the Messages API,
input_tokens does not include the cached input tokens, as
in Anthropic’s API.
Estimate usage cost
Calculate the usage cost of a request as:$0.002916:
$0.00360.
The response usage object reports the token counts needed for this
calculation. Account-level totals are available in the Token Factory
dashboard and from GET /api/v1/usage.