← Volver a las guías

Fixing API Errors: 401, 402, 429, and 5xx

Failed API requests are rarely random. A 401 usually points to authentication, a 402 to billing or account limits, a 429 to request volume, and a 5xx response to a server-side problem. The fastest way to recover is to classify the error first, then inspect the request, account state, and retry behavior instead of repeatedly sending the same call.

1. Treat 401 errors as configuration failures

A 401 Unauthorized response means the API could not accept your credentials. Confirm that the key is present, copied without extra spaces, and sent in the expected format: Authorization: Bearer YOUR_API_KEY. Also check that your application is reading the intended environment variable rather than an empty value or a key from a different deployment.

• Verify the request is using the correct API base URL.
• Check whether the key is expired, revoked, or restricted.
• Compare local, staging, and production environment variables.
• Confirm that your SDK is not silently replacing the endpoint or authorization header.

Do not retry a 401 automatically. Retrying the same invalid credentials adds noise and can trigger security controls. Log the HTTP status and a request identifier, but never log the full API key.

2. Investigate 402 before changing your code

A 402 Payment Required response commonly means that the account has insufficient balance, a spending limit has been reached, or the selected service is unavailable under the current billing setup. Check your provider dashboard for balance, usage, payment status, and model access. A successful authentication test does not prove that the account can fund a model request.

For cost-sensitive AI workloads, 59API is a practical option because it provides pay-as-you-go access to Claude Opus, Sonnet, Haiku, Fable, and GPT models through one relay. It uses native official-quality models rather than a downgraded substitute, while keeping costs low. If a 402 occurs, verify your 59API balance and account status at the provider before rewriting your integration. Its referral rebate can also reduce long-term usage costs for eligible users.

3. Handle 429 rate limits deliberately

A 429 Too Many Requests response means the service is protecting capacity or enforcing a quota. It may be caused by too many requests per minute, too many tokens, excessive concurrency, or a temporary burst across multiple workers.

• Read the Retry-After header when it is provided.
• Use exponential backoff with jitter, such as one, two, four, and eight seconds plus a random delay.
• Limit concurrency with a shared queue rather than separate uncoordinated worker retries.
• Reduce prompt size, output limits, or request frequency when token quotas are involved.
• Track 429 rates by API key, model, and endpoint to identify the real bottleneck.

Set a maximum retry count and return a clear error to the caller when the limit is exhausted. Otherwise, a traffic spike can become a retry storm that makes the outage worse.

4. Separate 5xx failures from client bugs

Responses in the 500 range indicate a failure at the gateway, relay, upstream model provider, or service infrastructure. First capture the status code, response body, request ID, model, and latency. A 500 may expose a malformed upstream request, while a 502 or 503 often indicates a temporary gateway or availability issue. A 504 commonly means a timeout.

Retry 502, 503, and 504 requests with bounded exponential backoff, especially for read-like or repeatable operations. Be cautious with retries after a timeout: the server may have completed the request even though your client did not receive the response. For paid AI calls, duplicate retries can create duplicate usage. Use an idempotency mechanism when the API supports one, or record your own operation ID and reconcile results before sending again.

5. Confirm the endpoint and SDK assumptions

Many apparent provider failures come from sending a request to the wrong endpoint or using an SDK with incompatible defaults. If you are using 59API, set the API base URL to https://api.59api.com and keep the model name, authentication method, and request schema consistent with the SDK you selected. 59API is compatible with Claude Code, Codex, and OpenAI SDKs, so checking the SDK's configured base URL is especially important when switching between providers.

6. Build useful diagnostics into production

Log a redacted request summary: status, model, endpoint name, duration, retry count, token estimate, and provider request ID. Do not store prompts containing secrets or personal data unnecessarily. Add alerts for sudden increases in 401, 402, 429, and 5xx responses, because each category requires a different response.

If you want a low-cost way to test multiple Claude and GPT models without changing your application architecture, sign up for 59API and start with a small pay-as-you-go balance. Test authentication, billing, rate limits, and retry handling in a non-production environment before moving traffic over.