Qwen3.8-Flash is QwenCloud’s hosted model with API ID qwen3.8-flash. To use it with a signed-in website, run Qwen Code locally, make the Tabbit skill available, and let the local agent use Tabbit CLI. The API model does not receive your browser cookies. QwenCloud lists function calling and regional availability in its model guide, and the Qwen Code setup shows the exact model ID. This is distinct from the open-weight Qwen3.8-Flash-Next checkpoint and the separate Qwen3.8-Omni-Flash route.
Key takeaways
Use the hosted QwenCloud ID qwen3.8-flash in a locally running Qwen Code client.
Qwen3.8-Flash-Next weights and Qwen3.8-Omni-Flash are separate routes.
Qwen Code skill discovery, Tabbit CLI reachability, endpoint availability, and website login are independent checks.
Ask for a named source-linked draft and stop before sharing.
Pick the Qwen route
| Route | What it is | Choice |
|---|---|---|
| Qwen3.8-Flash | Hosted QwenCloud API, ID qwen3.8-flash | Use in Qwen Code when this endpoint is available |
| Qwen3.8-Flash-Next | Open-weight checkpoint with separate local serving | Use only after checking hardware and parser |
| Qwen3.8-Omni-Flash | Separate multimodal route | Use when Omni inputs are needed, not as the Flash API alias |
Qwen Code documents the QwenCloud provider route. Its skills documentation says the client reads personal skills from ~/.agents/skills/ and supports /skills and /skill:tabbit. Verify tabbit appears before use. Then run tabbit-cli diagnose on the same machine. Skill discovery does not prove CLI reachability, and neither proves the website is signed in. Install Tabbit Browser, sign in yourself, and let the agent use a separate task tab so your other tabs remain available.
Draft a shared note
Use QwenCloud qwen3.8-flash in this local Qwen Code session. Use Tabbit CLI to read only the three team pages I authorize, in a separate task tab. Create a draft named "Qwen Flash launch brief, draft 01". Add the exact URL and a short evidence excerpt for each factual claim. Separate fact from inference, mark missing information unknown, and do not edit source pages. Stop before sharing or publishing and show me the draft for review.Expected output: a short brief with URL and evidence excerpt for each factual line, followed by unknowns. This is an example prompt, not an observed Qwen result.
Review before sharing
| Check | Pass | Stop |
|---|---|---|
| Model | QwenCloud route shows qwen3.8-flash | Wrong model or region |
| Browser | Expected site and account appear in task tab | Login page or wrong profile |
| Evidence | Each fact has a URL and excerpt | Unsupported paraphrase |
| Draft | Named draft remains unshared | Agent submits or publishes |
A Reddit user reported one Qwen Code run using hosted Qwen3.8-Flash that produced 74,558 output tokens over 11 minutes 25 seconds for a coding task; they said the final code passed syntax checking. This is one report, not a typical browser-task result. Another discussion describes account-switching loops with Qwen3.8 27B, which is a different route and not evidence about hosted Flash (post, 27B thread).
Tabbit as the local page layer
Tabbit Browser lets you review the signed-in page while the local agent works in a separate task tab. A fixture checked for this batch verified session reuse by a second Tabbit CLI task. It did not call QwenCloud or test an external model integration.
Install Tabbit Browser. Continue with the Tabbit CLI guide, existing browser connection, session reuse guide, first task checklist, MCP vs CLI, profiles and task tabs, Codex setup, and browser-agent security.
Choose this route after checking QwenCloud model and region. If the agent client runs remotely, verify a supported local handoff. The hosted model cannot reach a desktop browser on its own.
FAQ
What is the model ID?
Use the publisher’s exact model ID; third-party route names may differ. Check provider and region before sending a request.
Can the model open signed-in pages by itself?
No. A local agent must connect it to Tabbit CLI or another browser tool and run where the session is available.
How do I verify tool calling?
Check the model parser, inference server and agent client. The model card alone does not prove the host can execute browser tools.
How should I scope the first task?
Use a few authorized pages, read-only access, explicit fields, and a named draft for review.
Was this model tested with Tabbit?
No model-to-Tabbit run was completed for this draft. The local fixture only verified the browser session path.