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
- Aria's October 9 notice scheduled selected legacy Qwen models for retirement at 16:00 UTC that day. The date has passed, but the post alone does not establish the result for every account.
- A search for qwen in the public pricing catalog on November 4 returned 31 entries; the snapshot did not include the exact IDs qwen-turbo, qwen-vl-max, qwen3-8b, or qwen3-max.
- The catalog showed newer names such as qwen3.8-max, but the notice and catalog checked provided no one-to-one mapping from the older IDs.
- Check deployed configuration, API-key access, and task requirements one by one. Do not rely on a blind string replacement or automatic retries.
Read the October notice as a historical plan
Aria's official account posted a model-availability notice on October 9, 2026. It said selected legacy Qwen models were scheduled to retire at 16:00 UTC that day, or midnight on October 10 in China Standard Time. It named qwen-turbo, qwen-vl-max, qwen3-8b, and qwen3-max using the word “including,” so those four names were not presented as a complete list.
The notice concerned model availability in Aria's service. It was not an announcement that Alibaba Cloud or the official Qwen API was shutting down the Qwen family. The scheduled time has passed, but the post described a plan; by itself, it does not prove how retirement was applied to every account, route, or existing API key.
What the public catalog showed on November 4
When the Aria public pricing page was checked on November 4, 2026, it displayed 337 models. Searching for qwen returned 31 matching entries across two pages. The four exact IDs from the notice were not present in that dated public search snapshot. This was a catalog observation, not a test against an API endpoint or a specific account.
The same catalog showed newer Qwen names, including qwen3.8-max and qwen3.7-plus. It did not say which older model, if any, each one replaces, or promise identical input modalities, tool use, context, prices, or output behavior. Similar names are a reason to investigate further, not proof of compatibility.
Find every active use of the exact IDs
Search code, deployment configuration, model selectors, scheduled jobs, batch processes, and retry settings that you are authorized to inspect for each exact ID from the notice. Separate active settings from old documentation, samples, and archived records. For every active match, confirm whether it calls Aria or a different provider; do not replace every matching string across a repository.
Record the owner, task, deployed version, exact model ID, and the API key's applicable permissions for each workflow. Do not copy keys, customer prompts, or sensitive logs into a public support ticket. If the catalog and an existing call disagree, retain the timestamp, requested model ID, and a redacted error so Aria support can check the account's route.
Choose a candidate by task capability and price
Use Aria's current catalog and Model API documentation to shortlist candidates, then compare them against the inputs and outputs your task needs. Chat, vision, video, and embedding workflows are not interchangeable just because each model belongs to Qwen. Check support for tool calls, structured output, streaming, and any parameters your implementation depends on.
The retirement notice did not publish a replacement map, account eligibility, regional conditions, or price guarantees. The public catalog lists current models and billing information, but it does not guarantee that every route is available to every account. Before testing, review the price and limits that apply to your account, then compare representative, non-sensitive inputs for result quality, error handling, latency, and cost.
Roll out workflow changes in stages
Pin the candidate's exact ID in a reversible configuration copy and verify the value that your deployment system will actually ship. Test with limited traffic across normal responses, timeouts, rate limits, and the modalities your task uses. Changing a model in a web selector does not necessarily update a background script, proxy service, or retry job.
Expand usage only after quality, output format, cost, and reliability meet the task's needs. Set bounded retries and clear alerts; do not retry an old ID forever or silently switch to an unevaluated model after a failure. Keep the previous configuration available for rollback, but do not assume the old model will still be callable when you roll back.
Sources
Frequently Asked Questions
Can I rerun a scheduled task whose outcome is unknown during migration?
First inspect the original task’s logs, saved outputs and external-system receipts. Separate completed work, work confirmed not to have run and unknown outcomes. Changing a model does not prove that the earlier task made no writes or sent nothing. Resume only work confirmed unfinished to avoid repeating an action.
Does maintenance mentioned for other models in the October post mean they are still unavailable?
No. The post described wan3.0-video and gemini-embedding-2-preview as temporarily unavailable at that time, without a return date. That historical October statement does not establish their November 4 status or current availability to an account. Check the current catalog and applicable entry point separately.