Skip to content

Error Logs

Error logs help identify why a specific request failed. When troubleshooting, check Group Health first to understand the impact scope, then use the request_id in the error log to locate the exact request.

Entry: Console -> Usage Logs. Switch the log type to Error logs, then filter by time range, model, token, group, request ID, error message, or status code.

Quick checks

  • request_id locates one specific request
  • status_code groups errors by type
  • content searches for specific error keywords
  • group shows whether failures are concentrated in one group

Fields

FieldMeaningTroubleshooting value
TimeWhen the request failedCheck whether the issue is ongoing
ModelModel used by the requestCheck whether only one model is affected
TokenAPI key or token name used by the requestCheck whether only one token is affected
GroupPlan, model, or upstream resource group for the requestCheck whether failures are concentrated in one group
Request IDUnique identifier for one request, usually request_idProvide this first when contacting support
Error messageFailure reason text, usually containing contentSearch by keyword to find the exact error
Status codeHTTP or upstream status code, such as 401, 429, or 502Classify the error quickly

Troubleshooting Order

  1. Copy the request_id from the API response or log entry and search for that exact request first.
  2. If there is no request_id, narrow the scope by time range, model, and token.
  3. Group errors by status code, such as 401, 413, 429, 502, or 503.
  4. Search error message keywords, such as Invalid API key, daily limit exceeded, or Upstream request failed.
  5. If the same error is concentrated in one group, open Group Health to confirm the impact scope.

Attribution

Error logs help determine whether an issue is caused by usage, upstream services, or the platform. Attribution should not rely on status code alone; combine it with the impact scope.

AttributionHow to identify itTypical signsFirst action
Usage issueAffects only one request, one token, or a specific client configurationInvalid request body, parameter mismatch, copied token error, oversized context, wrong endpoint pathCheck local configuration, request body, model name, token, and tool version
Upstream issueMultiple requests show upstream failures, gateway timeouts, or upstream quota limitsUpstream request failed, all upstreams failed, bad response status code 502/524, upstream account rate limitsRetry later, reduce concurrency, or switch model/group if needed
Platform issuePlatform nodes, resource pools, or channels are unavailablesystem disk overloaded, No available accounts, No available channel, platform-side Service UnavailableRetry later; if it persists, contact support with the group, time range, and request_id
Needs group contextOne log entry is not enough to identify responsibilityErrors such as 429, 502, or 503 can have multiple causesCheck Group Health to confirm whether failures are concentrated

Note

Do not assign responsibility from a single 502 or 503. One failed request may be a transient upstream fluctuation. A large number of failures in the same group and time window is a better signal for upstream or platform resource issues.

Status Code Categories

Status codeCommon meaningInitial attributionFirst action
400Request body, parameters, or tool messages are invalidUsage issueCheck the client request body, endpoint path, and tool schema
401Token is invalid, disabled, or failed authenticationUsage issueCopy the token again and confirm its status
403Upstream rejected access, often due to permission or model access limitsUsage or upstream issueCheck plan, group, and model permissions
413Request body is too largeUsage issueShorten context, split files, or compress images
429Rate limit, daily quota limit, or credential cooldownNeeds group contextReduce concurrency and wait for quota or cooldown recovery
500Internal error from relay or wrapped upstream errorDepends on error messageRead the error message for the real cause
502Upstream service or intermediate network errorUpstream or network issueRetry later; switch model/group if needed
503Service, channel, account, or platform resource is temporarily unavailableUpstream or platform issueRetry later; contact support if it persists
504 / 521 / 522 / 524Gateway connection, read, or response timeoutUpstream or network issueReduce long-running requests and retry later

Common Error Messages

Error log contentMeaningInitial attributionWhat to do
status_code=401, Invalid tokenToken is invalid, copied incorrectly, or expiredUsage issueCopy the console token again and remove extra spaces
status_code=401, Invalid API key or key is pending admin approvalAPI key is invalid, or a new key is still waiting for approval/activationUsage issue or account statusConfirm you are using the latest console token; if it was just created or the plan changed, wait for it to take effect and retry. Contact support if it still fails
status_code=403, bad response status code 403Upstream rejected the request, commonly due to permission, account status, or model access limitsUsage or upstream issueCheck token, plan group, and model permissions; if it persists, switch models or contact support with the request_id
status_code=413, openai_error / bad response status code 413Request body is too large, often due to context, files, images, or tool resultsUsage issueShorten context, reduce one-time uploads, split large files, or compress images before retrying
status_code=429, Account RPM limit exceededThe upstream account hit its per-minute request limitUsage issue or upstream limitReduce concurrency and retry frequency, then try again later
status_code=429, Account daily limit exceededThe upstream account hit its daily request limitUpstream limitWait for daily quota reset, or switch to another available model/group
status_code=429, All credentials for model ... are cooling downAll upstream credentials for the current model are cooling downUpstream limitReduce retry frequency and wait for cooldown; switch models temporarily for urgent tasks
status_code=500, 请求失败 [429]: {"message":"Too many requests, please wait before trying again."}Upstream returned a 429 rate limit, but the relay wrapped it as 500Usage issue or upstream limitTreat it as rate limiting: reduce concurrency, increase retry intervals, and avoid immediate repeated retries
status_code=500, auth_unavailable: no auth availableNo upstream authentication resource is currently available for the model or groupPlatform issueRetry later; if it persists, switch to an available model/group or contact support to confirm plan and group status
status_code=502, Upstream request failed / bad response status code 502 / all upstreams failedUpstream service or intermediate network returned an error, or all available upstreams failedUpstream issueRetry later; if it persists, switch models or contact support with the request_id
status_code=502, openai_errorAn OpenAI-compatible upstream returned an error, but the relay did not receive a more specific reasonUpstream issueTreat it as an upstream error first; if the same request reproduces consistently, reduce context and provide the request_id to support
status_code=502, The origin web server returned an invalid or incomplete response to CloudflareThe upstream origin returned an invalid response through CloudflareUpstream issueUsually a temporary upstream failure; retry later
status_code=500, upstream error: do request failedThe request failed while being sent upstream, often due to network connectivity or temporary upstream unavailabilityUpstream or network issueRetry later; if it persists, provide the request_id to support
status_code=520, bad response status code 520Cloudflare returned an unknown error, usually due to an abnormal upstream response or interrupted connectionUpstream or network issueRetry later; if concentrated, treat it as an upstream incident
status_code=521 / 522 / 504 / 524Cloudflare or upstream gateway connection, read, or response timeoutUpstream or network issueRetry later first; for long-running tasks, reduce context, output length, or tool-call chains
status_code=503, Service UnavailableUpstream service is temporarily unavailable, or the current group has no available resourcesNeeds group contextRetry later; if it persists, switch model/group or contact support to confirm resource status
status_code=503, system disk overloadedA service node has high disk usage, so the platform temporarily rejects requests to protect service stabilityPlatform issueThis is platform-side capacity or node state; retry later and contact support if it persists
status_code=503, No available accounts: no available accountsThe current group has no available upstream accountPlatform issueSwitch model/group or retry later; if it lasts for a long time, contact support to confirm the resource pool
status_code=503, No available channel for model ... under group ...There is no available channel for this model under the current groupUsage or platform issueCheck the model name and plan support scope, then switch to a model supported by this group
status_code=503, model gpt-image-2 is only supported on /v1/images/generations and /v1/images/editsAn image model was called through the wrong endpointUsage issueSend image generation/edit requests to the corresponding images endpoint
status_code=500, Image source is a local path that is not readable from this serverThe request contains a local image path that the upstream cannot read, which may leave terminal input unresponsiveUsage issueFor frontend projects, first check lock dependency files: delete related lock files, or remove abnormal png fields from them and reopen the session. To keep sending images, use a public http(s) image URL or a data:image/... base64 payload
status_code=500, failed to parse multipart formThe image or file upload request body does not match the API formatUsage issueCheck multipart/form-data field names, file fields, and request headers; do not write an invalid boundary manually
status_code=400, Invalid request: prompt is required / 解析 Images 请求失败: prompt 不能为空Image generation/edit request is missing promptUsage issueProvide a non-empty prompt and confirm the request is sent to an image endpoint
status_code=400, Unsupported parameter: messagesEndpoint and parameter format do not match, commonly when Chat/Responses parameters are sent to an endpoint that does not support messagesUsage issueCheck the tool base_url, endpoint path, and model type; adjust the request body for the target API
tool_use ids were found without tool_result blocks immediately afterTool-call message sequence does not meet the Claude protocol requirementsUsage issueKeep each tool_use immediately followed by its corresponding tool_result; if generated by a tool, upgrade the client or start a new session
Invalid schema for function ... None is not of type 'array'Tool function schema is invalid, usually because parameters or an array field is empty or has the wrong typeUsage issueCheck MCP/tool definitions, provide [] for array fields, and ensure function parameter schemas follow JSON Schema
status_code=500, not implementedThe current endpoint, model capability, or tool-call path is not implementedUsage or platform issueConfirm you are using a supported endpoint and model; switch models or contact support if needed

Handle by Type

Authentication and permissions

Focus on 401, 403, Invalid API key, pending admin approval, and No available channel. First confirm that the token comes from the console, was copied completely, and has permission for the current model and group.

Request format

Focus on 400, 413, Unsupported parameter, prompt is required, and Invalid schema. First check the endpoint path, request body fields, tool schema, image request parameters, and context size.

Rate limits and quota

Focus on 429, Max 10/min, daily limit exceeded, and cooling down. Do not retry with no delay. Reduce concurrency according to the log message, or wait for quota/credential recovery.

Upstream and platform resources

Focus on 502, 503, 504, 520, 521, 522, 524, Service Unavailable, and system disk overloaded. Retry later first; if errors are concentrated in the same group, check Group Health.

Information for Support

For simple issues, check Error Logs and Group Health first. If the issue remains unresolved, open the error log details in console/log and click the copy icon to copy the troubleshooting details in one click. When contacting support, provide the following in one message so the technical team can investigate with less back-and-forth:

  • User ID
  • Time range: when the issue started and when it last appeared
  • Group name: group
  • Model name used by the request
  • Status code, such as 429, 413, 502, or 503
  • Error content: error_reasons.content
  • Request ID: request_id from a single log entry or API response
  • Impact scope: one token, one model, one group, or multiple groups at the same time

Summary

Error logs explain why one request failed. Group health shows whether the issue is concentrated. Use both together for faster troubleshooting.