Service eligibility and regional restrictions
PuppyIP serves only compliant overseas businesses and their authorized personnel. Proxy services are not available in mainland China. The service may only be used for lawful business activities outside mainland China. Use of this service within mainland China is prohibited.
Hosting a proxy IP or server overseas does not change these restrictions. The service must not be provided to end users in mainland China through relaying, forwarding, sharing or resale. Before use, read the Terms of Service.
Key Takeaways
- Incident alzmexlqangd moved to resolved at 09:37:03.607 UTC on October 15. The reviewed record also shows the Workers AI component as operational.
- The announcement named two specific DeepSeek V4 Flash and V4 Pro versions, plus GLM-5.2, GLM-5.3 and GLM-5.3-Flash. It does not establish an impact on every Workers AI model.
- The investigation update appeared at 07:10 UTC, monitoring followed at 07:57 UTC, and resolution was announced at 09:37 UTC. Monitoring and resolution are separate stages.
- Compare model IDs, endpoints, timestamps and error details from existing logs. The announcement supplies no universal error code, final repair details or customer retry policy.
Resolved at 09:37 UTC: where does the monitoring update fit?
Cloudflare's official incident, Cloudflare Workers AI Issues, is recorded as resolved. The latest update was published at 09:37:03.607 UTC on October 15, 2026, or 17:37:03.607 in Asia/Shanghai. It supersedes the earlier monitoring status.
At 07:10:58.804 UTC that day, Cloudflare described possible inference-request errors for certain models and said work was under way. The update at 07:57:49.667 UTC said monitoring was continuing; resolution followed at 09:37:03.607 UTC. The first announcement is not a start time for every customer's errors, and the closing update provides no account-by-account recovery results.
Match the five full model IDs first
The two DeepSeek IDs are @cf/deepseek-ai/deepseek-v4-flash-0731 and @cf/deepseek-ai/deepseek-v4-pro-0813. Their suffixes are part of the exact identifiers in the announcement. A screen showing only the DeepSeek family name does not establish which version a request used.
The three GLM IDs are @cf/zai-org/glm-5.2, @cf/zai-org/glm-5.3 and @cf/zai-org/glm-5.3-flash. Check which ID appears in your existing request records before matching it to the list. Similar model names on another platform do not automatically fall within this incident's scope.
The announcement does not say every model was affected, identify affected regions or give a share of customers impacted. Nor does it establish the current health of models outside that list. It describes possible inference-request errors. It does not announce a change to model eligibility, pricing or account quotas.
Place existing errors in the correct time window
The following is editorial advice for organizing an investigation. Start with the actual request time in existing application logs, monitoring or support records. Label its time zone and align it with the three UTC milestones above. Record ticket submission and screenshot times separately; neither should replace the request timestamp.
For a hypothetical example, an error message arrives at 17:40 in Asia/Shanghai, but the request failed at 17:30. That request occurred at 09:30 UTC, during the monitoring stage. A message arriving after the resolution notice does not mean the failed request also occurred after it. This example explains time alignment; it establishes no cause for a real failure.
What should you retain if requests fail after resolution?
For the same failed request, collect the full model ID, endpoint, UTC timestamp, recorded response status and error details. Include a request identifier if your existing records contain one. Keep these fields tied to a single request rather than combining failures from different models or times into one conclusion.
If existing records show failures after the resolved timestamp, continue investigating those specific requests. You can provide sanitized evidence through your existing support channel. Mark missing fields as unknown. One failed request does not establish that the official incident has reopened, and a closed status entry cannot determine whether every request in your application succeeded.
Which repair details remain unconfirmed?
The three updates provide no root cause, final repair method, universal HTTP error code, retry interval or customer command. They therefore do not establish that changing an IP address, model, authentication setting or network configuration will resolve continuing errors. Preserve the existing evidence that can be matched to a specific request before deciding what to do next.
The reviewed incident record labels the impact minor and shows the Workers AI component as operational. These are official status fields, not availability measurements across every account. No model inference, connectivity test or configuration change was performed for this article, and it does not verify recovery for each application.
Sources
Frequently Asked Questions
Can I match the incident if my application records only a model alias?
Use existing configuration or request records to establish the full Workers AI model ID that the alias represented at the time. A family name or similar display label cannot identify the exact version. If the mapping is unavailable, leave the match unknown rather than treating the model as included or excluded.
Do continuing errors mean I need a paid upgrade or new model access?
This incident announcement introduces no plan, price, quota or access-permission change. Retain the specific response before drawing a conclusion. Neither the incident nor its resolved status establishes that an account needs an upgrade, or guarantees that an account has no separate permission issue.
Should I include an API key or complete input when contacting support?
Start with the minimum information needed to locate the request, such as the model ID, timestamp, request identifier and sanitized error details. Do not proactively include an API key or unrelated full input. Follow the existing support channel's specific instructions if more information is needed.