Gemini 3.5 Flash is no longer selectable in Google AI Studio. Google says API requests that use gemini-3.5-flash are automatically routed to Gemini 3.6 Flash. For signed-in pages, verify the model route and local agent first. Then use Tabbit Browser with Tabbit CLI to read a small, authorized set of pages and stop at a named draft for review. The browser connection is a separate local tool; Gemini does not inherit your desktop session.
A Reddit browser-agent user recommends pausing before the final external action when session state and result checks are uncertain. That is workflow feedback, not evidence about Gemini. Original discussion.
Key takeaways
Treat
gemini-3.5-flashas a deprecated request ID. Google routes it togemini-3.6-flashand recommends changing the model string.gemini-3.5-flash-liteis a separate model and lifecycle, not a renamed full Flash model.Gemini API and Google Cloud Agent Platform publish different lifecycle information. Check the service you call.
Verify model availability,
tabbitskill discovery, CLI reachability, and website sign-in separately.Compare records by stable ID, keep source URLs beside values, and stop before editing or submitting either page.
Separate the model, client, and browser
| Layer | Verified status | Check before starting |
|---|---|---|
| Gemini API | gemini-3.5-flash is deprecated and requests route to 3.6 Flash | Set a current model ID in the client; confirm its route |
| Gemini 3.5 Flash-Lite | Separate ID: gemini-3.5-flash-lite | Choose it intentionally; do not treat it as full Flash |
| Local agent client | Local tools work only if the host permits them | Check provider, command permissions, and skill discovery |
| Tabbit CLI | Connects a local task to a Tabbit browser environment | Diagnose the CLI and confirm the selected instance |
| Website session | Pages appear as signed in only in the chosen profile | Open an authorized page and confirm it did not redirect to login |
Discover the installed tabbit skill with the client’s documented mechanism. Some clients use /tabbit; Pi uses /skill:tabbit. Skill discovery only confirms that the instructions loaded. It does not prove the CLI can reach the intended browser or that the website is signed in. See the Tabbit CLI guide, browser connection walkthrough, and first-task checklist.
Check the correct lifecycle
The Gemini Developer API model page marks gemini-3.5-flash deprecated and says it is no longer selectable in AI Studio. The endpoint is not shutting down because API requests are automatically routed to gemini-3.6-flash. Google recommends updating to gemini-3.6-flash or gemini-3.8-flash. Read the Gemini 3.5 Flash model page and Gemini API deprecations.
Google Cloud lists its own lifecycle for Gemini Enterprise Agent Platform. There, the retirement date for gemini-3.5-flash is May 19, 2027 or later. That Cloud date does not replace the Gemini API deprecation or describe AI Studio routing. If your client uses Cloud, check the actual project in the Cloud model lifecycle table.
Do not silently substitute Flash-Lite or a newer Flash model. The API lists gemini-3.5-flash-lite separately and currently gives it no shutdown date. Gemini 3.6 and 3.8 Flash have separate IDs too. Make any migration explicit, then recheck the browser task. See the Gemini 3.6 Flash guide and Gemini 3.8 Flash guide.
Reconcile one record across two signed-in pages
Use a read-only comparison, such as matching a dashboard item with its source record in another authorized system. Give the agent a stable ID and ask for the title, displayed status, visible update date, and page URL from each site. A Reddit user said session state, page changes, and checking whether a submission landed were less reliable than the model understanding the request. They advised pausing before an external action and reopening the result page. This is one person's experience, not Gemini evidence. Original thread.
| Source | Record | Rule |
|---|---|---|
| Signed-in dashboard | Stable ID, displayed status, visible date, URL | Match the exact ID, not a similar title |
| Signed-in source page | Same ID, source status, visible date, URL | Flag conflicts and preserve each site's wording |
| Draft | Gemini browser cross-check — draft 01 | Add one source URL per value and unresolved fields; do not write back |
Use this prompt from a local agent that can run Tabbit CLI. Replace the example pages and ID with ones you are authorized to read:
Use the installed tabbit skill and Tabbit CLI from this local agent. Read only these two pages in separate task tabs: [dashboard URL] and [source record URL]. Compare stable record ID REC-204. Return a table with field, dashboard value, source-page value, URL for each value, and status (match, conflict, or unknown). Do not edit either page. Prepare a browser-visible draft titled "Gemini browser cross-check — draft 01" with unresolved fields. Stop for human review before saving, sending, or sharing. If the model route, local CLI, browser connection, or site login fails, name that gate and return a source checklist.A second Reddit discussion notes that a page can load stale or partial state. One commenter recommends reading the page again after each step instead of trusting a click. Treat this as a review habit, not a model capability claim. Original thread.
Run it with Tabbit Browser
Install Tabbit Browser and follow the Tabbit CLI quick start. The task tab is separate, so you can continue working in your usual browser. A local project fixture verified session reuse across two named CLI tasks; it did not test Gemini access or a Google–Tabbit integration.
A cloud client cannot reach a desktop session without a supported local handoff. Ask it for a source plan or draft, then run the browser step locally. For setup and boundaries, see Browser MCP vs CLI, local vs cloud browser agents, research with source links, and safe login reuse.
FAQ
Can I select Gemini 3.5 Flash in Google AI Studio?
No. Google's current model page says it is no longer selectable there. API requests using the old ID are automatically routed to Gemini 3.6 Flash.
Is Gemini 3.5 Flash-Lite the same model?
No. Flash-Lite is a separate model ID with its own lifecycle. It is not the full Flash model under a new name.
Does Google Cloud's retirement date apply to Gemini API?
No. Google Cloud publishes a separate lifecycle for its Agent Platform. Check the service and project you actually use.
Can Gemini 3.5 Flash access my signed-in browser by itself?
No. A model endpoint is not a desktop browser connection. Configure a local agent client and Tabbit CLI separately.
What if the client runs remotely?
Ask it for a source plan or draft, then hand the browser step to a local agent that can reach Tabbit CLI. Review before sharing.