Troubleshooting & help

Error codes

Find the cause by HTTP status code, with a self-check list.

Check the HTTP status code first, then the error message in the response body. Most tools display both.

`400`Request violates the usage policy

Symptom: 400 with the message: Anthropic refused this request under its usage policy. Do not retry this request.
Common causes: The request content violates Anthropic’s usage policy and was refused upstream.
What to do: Do not retry the request; repeated violations get your device banned. Keep your usage within the policy. If you have been banned, open a support ticket to request an unban.
Billing: Refer to the usage records in the console.

Symptom: 401 Unauthorized, invalid API key, or authentication failed.
Common causes: An incomplete or deleted key, a missing Bearer prefix for OpenAI requests, or a key entered in the wrong protocol field.
What to do: Copy the key again, check the authentication header, and create a replacement key if needed.
Billing: Not billed.

Symptom: 403 or an insufficient balance, quota, or subscription message. The group may also deny access to the model.
Common causes: The balance is depleted, the plan quota has not reset, or the key’s group does not support the model or protocol.
What to do: Check balance and usage in the console. Top up, wait for quota renewal, or use a key bound to the appropriate group.
Billing: Requests that are not executed are not billed.

Symptom: 404 Not Found or model not found.
Common causes: An extra or missing /v1, using /v1 for Gemini, or an incorrect or unavailable model name.
What to do: Check the Base URL guide and copy the exact model name from the catalog.
Billing: Not billed.

Symptom: 413 Payload Too Large.
Common causes: Context, attachments, or conversation history exceed the limit.
What to do: Reduce context, remove unnecessary files, or start a new conversation.
Billing: Not billed.

Symptom: 429 Too Many Requests; the tool may stop connecting or repeatedly fail.
Common causes: Too much concurrency, multiple active tools or sessions, or retries that are too frequent.
What to do: Reduce concurrency, close extra clients, and wait before retrying.
Billing: Incomplete requests are not billed.

Symptom: 500 / 502 / 503 / 504.
Common causes: Upstream service disruption, network instability, or temporary maintenance.
What to do: Retry later or try another model. Open a support ticket if the issue persists.
Billing: Billing follows actual usage recorded in the console.

Is the key complete and still active in the console’s API keys list?
Does the Base URL match the protocol? See the tool reference on the Base URL page.
Is the model name listed in the Model catalog with exactly the same capitalization?
Does the key’s group support the protocol you are using?
Is there enough balance or subscription quota?
Are several tools or sessions sending requests concurrently?
Did you fully restart the tool after changing its configuration?

If these checks do not resolve the issue, open a support ticket in the console and include the time, model name, status code, and response body.