Docs / Getting started
Billing & discounts
Settled per token in real time at each model’s published discounted price, quoted in USD — no monthly fee, no tier thresholds.
Concept
Description
Model discount
Each model carries its own discount, with no account tiers. The struck-through figure on the pricing page is the upstream list price; the one beside it is what you are charged.
Cache hit
Input tokens that hit the prefix cache are billed at 20%; output is never discounted.
Settlement timing
Settled immediately when the request completes; streaming requests are billed on the final usage.
Failed requests
4xx responses and 5xx responses where upstream produced no tokens are not billed.
A worked example
Take DeepSeek-V4 (input $0.28 / output $0.42 per million tokens). A request with 8,000 input + 1,200 output tokens:
TEXT
Input 8,000 / 1e6 × 0.28 = $0.00224 Output 1,200 / 1e6 × 0.42 = $0.00050 Total = $0.00274
Trial credit is valid for 30 days; purchased credit and referral commission never expire. When the balance runs out, the API returns 402.