# Errors

Errors use the OpenAI shape, so a client that already handles them needs no change.

| Status | Code | What to do |
|---|---|---|
| 401 | `missing_api_key`, `invalid_api_key` | Check the `Authorization` header. A revoked key, an expired key and a suspended account all answer the same way. |
| 402 | `insufficient_credit` | Top up. `X-Zanii-Topup-Url` points at the console. |
| 403 | `model_not_allowed` | This key is restricted to other models. |
| 402 | `spend_cap_reached` | The account passed its monthly ceiling. Raise it in the console. |
| 409 | `idempotent_replay` | This request id was already served. Send a new one. |
| 501 | `embeddings_not_supported` | This deployment's model server has no embedding model. |
| 501 | `not_supported_by_model_server` | This deployment's model server does not serve that route. The message names what to use instead. |
| 503 | `tokenizer_unavailable` | The model server cannot count tokens right now. |
| 404 | `model_not_found` | The model is not in the catalogue. |
| 429 | `rate_limit_exceeded`, `too_many_concurrent_requests` | Back off; `Retry-After` says how long. |
| 502 | `upstream_error` | The model server failed. The call is not charged. |
| 503 | `receipts_unavailable` | The key requires a receipt and none could be signed. The call is not charged. |

A refused call never reaches the model, and is never billed.