Billing
Each call is priced from the tokens the provider reports, on four separate meters.
Four meters
- Input, output, cache write and cache read each have their own rate per 1M tokens.
- Cached input reported by the provider is moved out of input and billed at the cache-read rate.
- If a successful call reports no usage at all, it is billed as 2,000 input + 2,000 output tokens and marked as an estimate.
Balance and HTTP 402
Pay-as-you-go keys need a balance above zero. A call that finishes as the balance runs out can take it down to −¥1 at most; after that calls answer 402 until you top up.
Monthly plans for heavy agent weeks
Monthly plan metering, step by step
- Each call is valued at the official list price in official USD, multiplied by the plan's multiplier, and that amount is drawn from the quota.
- Claude plans cover Claude models only; GPT plans cover GPT models only.
- A plan is spent through a dedicated key, created automatically at purchase. Pay-as-you-go keys never draw on a plan.
- Once the quota or the weekly ceiling is used up, the dedicated key answers HTTP 402. It does not fall back to your balance.
- The weekly ceiling resets 7 days after the first draw that follows the previous reset, not on calendar weeks.
- Plans stack: every purchase is a separate grant, consumed in order of expiry. Quota left at expiry does not roll over.
- Gemini, DeepSeek, Kimi, GLM, Qwen, Grok, MiniMax, Hunyuan and MiMo are metered-only.
Referrals and withdrawals
Direct referrals earn 11% of each payment, the second tier 4%. Withdrawals go to Alipay from $16.67 and are reviewed by hand; earnings can also be moved to your balance.
Refunds
Unused balance from a top-up can be refunded within 8 days of it; the full rules are in the Terms