ERROR CODES / ENGLISH INDEX

The twenty error codes you will meet first

Short English summaries, with a link to the full Chinese page for each code. The code itself is language-independent; the detail pages are where the request examples and vendor-specific text live.

Code	Meaning
429	429 Too Many Requests Rate limit: wait, then back off with jitter.
401	401 Invalid API key The credential was not accepted; check whitespace and environment.
402	402 Insufficient quota Balance, quota window, or a key without quota.
402	402 Insufficient balance The account is out of credit; top up or switch keys.
403	403 Permission denied The key is valid but not allowed to use this model or group.
404	404 Model not found The model ID is wrong or not enabled for your group.
422	422 Invalid parameter A request field has the wrong type, name, or range.
500	500 Server error The upstream or relay failed; retry with a cap.
502	502 Bad gateway The relay could not reach its upstream.
504	504 Gateway timeout The upstream did not answer before the relay gave up.
529	529 Overloaded The upstream is saturated; route elsewhere or wait.
—	Context length exceeded The prompt plus output exceeds the model window.
—	Connection error DNS, proxy, TLS, or a dropped socket before a response.
—	TLS certificate error The certificate chain or hostname does not verify.
—	Streaming not supported The relay or model does not support SSE streaming.
—	Stream disconnected The SSE connection closed before the final event.
—	Token count mismatch Usage fields disagree with the visible text or price.
—	Suspected model swap Quality or tokenizer evidence suggests a different backend.
—	Slow first token Time to first token is high; check queueing and model size.
—	Timeout The client or gateway gave up before the response completed.

Looking for the full library? The Chinese 报错百科 covers 110 codes with request examples, cause rankings, and responsibility tables.