Alibaba Cloud Model Studio retired qwen3.6-max-preview on October 10, 2026 and lists qwen3.7-max as the replacement. OpenRouter’s endpoint catalog also returned no providers for the old preview on November 10. Use the saved Preview results as the baseline if you have them, then rerun the same read-only page with an available model and a local agent using Tabbit CLI. Alibaba Cloud retirement notice, OpenRouter endpoint catalog
This workflow reruns the same read-only page task after a model switch. The agent captures visible values and page evidence, then shows a named comparison draft in a separate Tabbit task tab. It does not write to the dashboard.
Key takeaways
Verify account, region, API model ID, and tool-call support separately.
Treat Max Preview, Plus, Flash, and open-weight releases as different model routes.
The original release took text input and produced text, so collect visible page text and links.
Use an archived run or confirm a separate exact route.
Use Tabbit CLI for local browser actions and keep a human review checkpoint.
| Control | Keep constant | Record |
|---|---|---|
| Page | Same authorized URL and account | URL and page title |
| Time | Same visible date range | Date range and timezone if shown |
| Provider and model ID | Record both before and after | Keep the account and region visible |
| Review | Same extraction prompt and criteria | Observed, inferred, or unknown |
Choose the Qwen route before connecting the browser
| Route | What it is | What to verify |
|---|---|---|
| Qwen3.6 Max Preview | Retired Model Studio ID qwen3.6-max-preview | Use an archived run or confirm a separate exact route |
| Qwen3.6 Plus | A separate hosted Qwen route | Its own model ID and plan/API access |
| Qwen3.6 Flash | A separate speed-oriented route | Availability and supported tools in your account |
| Qwen3.6 open-weight release | A downloadable model checkpoint | Exact checkpoint, license, runtime, hardware; it is not Max Preview |
| Tabbit CLI | Local browser executor | Target Tabbit environment and website login |
Re-run the same page without confusing the model routes
Start with an existing Preview result, or use the exact preview only if your provider can verify a callable route. Mark the provider and model ID for each run. The original release took text input and produced text, so collect visible page text and links.
The name “Max” can be confused with a product tier or another model family. Use the literal ID from the current provider console. Qwen Studio chat, a coding plan, and a Model Studio API key are separate access paths. A community question about whether Max Preview was included in a coding plan illustrates why account-level checks matter; it does not establish current availability for your account.
Connect the API client to Tabbit CLI
Start with an existing Preview result, or use the exact preview only if your provider can verify a callable route. Mark the provider and model ID for each run. The original release took text input and produced text, so collect visible page text and links.
Check the agent client. Confirm it accepts the selected API and permits local tools. API function calling and browser control are separate capabilities.
Install Tabbit Browser. Sign in to the authorized dashboard in the target Tabbit environment, then connect Tabbit CLI from a supported local agent. Supported clients discover /tabbit; Pi uses /skill:tabbit; some clients expose a /skills menu. Check skill discovery before proceeding.
Open a separate task tab. Keep the existing sign-in in the Tabbit environment. Start with a read-only page and collect URL, page title, visible time range, and result fields. A local fixture check recorded for this workflow showed a second task could read a test page after a first task established a short-lived login. That verifies Tabbit CLI session reuse only, not Qwen integration.
Prompt for a controlled before-and-after check
This is a test recipe, not a claim that a particular client has been connected to Qwen3.6 Max Preview. Confirm the route first.
Install Tabbit Browser and set up Tabbit CLI on this computer. Tabbit CLI quick start
Use Tabbit CLI in a new task tab to read the same authorized dashboard page. Do not change filters, submit, export, or edit. Capture the page title, source URL, visible date range, field labels, current values, and any visible caveats. Prepare a text draft named “Browser model comparison”. Mark each item as observed, inferred, or unknown. Compare it with the previous model run only where the page evidence is identical. Show both notes in the task tab and stop for human review.Keep the comparison narrow
A controlled rerun changes the model route while holding the browser page, date range, extraction request, and review criteria constant. Do not treat a difference as a model-quality result if the page changed or one run used a different provider, context setting, or tool adapter. Record those differences beside the draft.
On OpenCode, a user reported testing preserve_thinking with several Qwen3.6 variants and described the adapter requirements. Zhidongxi’s launch-day tests compared generated coding prototypes in Qwen Studio. Both reports concern the original release period; neither tests Tabbit or establishes current endpoint access. OpenCode user report, Zhidongxi hands-on report
Use Tabbit for the browser step, not the model switch
Install Tabbit Browser, keep the authorized site signed in there, and run Tabbit CLI from a local agent in a separate task tab. That lets you continue other work in the same browser environment while the agent gathers page evidence. The model provider handles inference; the local agent handles tool orchestration; Tabbit CLI handles browser actions. Confirm each layer independently, then inspect the named draft before reusing it.
When this setup fits
Use it for a short, read-only comparison where the same page can be revisited and a person will inspect the notes. It does not suit a production automation that depends on a preview remaining unchanged, or a task that needs visual screenshot understanding from a text-only model. Choose a stable route or a model with verified visual input if those requirements matter.
FAQ
Is Qwen3.6 Max Preview the same as Qwen3.6 Plus?
No. Max Preview is a separate hosted model route with its own model ID. Plus is a different route, and an open-weight Qwen3.6 model is not the hosted Max Preview.
What model ID should I use?
Retired Model Studio ID `qwen3.6-max-preview`. Use an archived run or confirm a separate exact route.
Can the model see screenshots?
The original release took text input and produced text, so collect visible page text and links.
Does Qwen Studio access mean my coding plan includes it?
No. Chat access, coding plans, and API access are separate products. Check the Model Studio console and plan documentation for your account.
Will preview lifecycle changes affect my setup?
Alibaba Cloud Model Studio retired `qwen3.6-max-preview` on October 10, 2026 and lists `qwen3.7-max` as the replacement.
Does the agent change the dashboard?
The example is read-only and stops at a named draft for review. Keep writes, filters, exports, and submissions outside the task until you approve them.