Errors & retries

The unified error format and how to back off.

Status codes

HTTP	Meaning	What to do
401	Key invalid or missing	Check the Authorization header
402	Balance exhausted	Top up, or claim the starter credit
403	Key not entitled to this model	Check the key's model scope in the console
429	Rate limited (RPM/TPM)	Retry with exponential backoff
5xx	Upstream model failure	The gateway cools down the bad channel and retries; you can also back off and retry

Retry example

import time
for attempt in range(4):
    try:
        resp = client.chat.completions.create(model="auto", messages=msgs)
        break
    except Exception as e:
        if attempt == 3: raise
        time.sleep(2 ** attempt)   # 1s 2s 4s