Billing Rules and Error Codes

The billing formula#

Per-token models:

cost = model base price × group multiplier × token usage

• Base price: the "USD per 1M tokens" on the pricing page, input and output priced separately
• Group multiplier: default = ×1.0; Claude official / OpenAI official / Gemini premium = ×1.5
• Token counts are converted with each model's tokenizer; expect small (±5%) differences from local estimates

Per-call models (image / video / 3D): billed at the flat "$/call" price on the pricing page.

Reconciling your bill in three steps#

1. Console → logs: find the call, confirm model, group and timestamp
2. Multiply usage (prompt_tokens / completion_tokens) by the pricing-page unit price
3. Compare with the charge record; mismatches usually mean a non-default group was used

Groups#

Error code reference#

Error body format:

{
  "error": {
    "message": "This token cannot use model xxx",
    "type": "forbidden"
  }
}

Retry template#

import time, random

def with_retry(fn, attempts=4):
    for i in range(attempts):
        try:
            return fn()
        except Exception as e:
            status = getattr(e, "status_code", None)
            if status in (429,) or (status and status >= 500):
                wait = (2 ** i) + random.random()
                time.sleep(wait)
                continue
            raise
    raise RuntimeError("retries exhausted")

Principle: only retry idempotent, potentially-recoverable errors (429, 5xx). 400/401/404 mean fix your configuration — retries are pointless.

Cost control checklist#

• Set max_tokens on every call
• Keep system prompt prefixes stable to hit prompt caching
• Prefer low-priced default-group models for high-frequency paths
• Watch the daily usage curve in the console; investigate spikes same-day

For billing disputes, use the channels on the contact page with timestamps, model names and log screenshots.