For a signed-in preview check that previously used Gemini 3.7 Flash, install Tabbit Browser and use Tabbit CLI from a compatible local agent. Google AI Studio has removed 3.7 Flash from its selector, and Gemini API requests to the old ID now run 3.8 Flash. Record the version your provider actually runs before comparing results.
Google Cloud’s Gemini Enterprise Agent Platform has a separate lifecycle. Its table lists January 28, 2027 as the retirement date for gemini-3.7-flash. Record which service ran the model. Google Cloud model lifecycle
Key takeaways
Google AI Studio removed
gemini-3.7-flashfrom its selector. Google’s Gemini API reroutes the old ID togemini-3.8-flash.Record the actual model in your run log. A request sent to the old ID on Google’s API now belongs in the 3.8 results.
A compatible local agent can connect to Tabbit Browser through Tabbit CLI, but that integration is separate from model availability.
Check the model route, CLI launcher, website login, and page permissions as separate layers.
Keep the first browser task read-only and review the cited fields before changing a record.
Check which model you are actually calling
| Layer | What to verify | Stop condition |
|---|---|---|
| Model ID | Does the client still expose gemini-3.7-flash? | The provider silently substitutes a different model |
| Current route | Does Google document routing to gemini-3.8-flash? | Logs do not reveal the actual model |
| Local tools | Can the client discover Tabbit skill and launch Tabbit CLI? | Skill is found but launcher cannot run |
| Website session | Is the correct account signed in within Tabbit? | Login redirect or wrong workspace |
Google’s Gemini 3.7 Flash model page describes the current routing. An earlier 3.8 launch post said 3.7 was still supported at the time. Use the current API documentation and the model metadata returned by your provider.
Connect a local agent to a signed-in page
Install Tabbit Browser and sign in to the approved preview site yourself. In a compatible local client, load the Tabbit skill with the syntax that client supports. Pi uses /skill:tabbit; some clients use /skills, and others discover installed skill files. Then confirm the client can start the local Tabbit CLI launcher. Selecting the model, discovering the skill, running the launcher, and signing in to the site are four separate checks.
A local task can use the existing Tabbit session in a separate task tab, leaving your working tabs available. This is a browser-session path; it does not guarantee that a specific client exposes Gemini 3.7 or that an external agent can reach your local browser. Remote runners need a documented local bridge. If the client does not offer 3.7, choose a currently listed model and record its actual ID.
Read a preview page without changing it
This illustrative workflow uses a staging preview the operator is allowed to access. It does not claim a live Gemini-to-Tabbit test. Ask the model to extract only the visible title, status, and URL; stop before it edits or publishes.
Use only the preview page open in my Tabbit task tab. Read visible text; do not click save, change settings, or follow links.
Create an unsent document draft named "Preview status check — Gemini 3.7 review". Include the page title, visible status, source URL, and any mismatch between the page and the brief. Mark absent fields as unknown.
Stop at a review checkpoint and show the complete draft with its source URL. Do not share or publish it.
Expected output: title, status, URL, mismatch list, and draft link.Review the page yourself and verify the exact status. A model response does not prove that it saw the right account or the intended preview. If the provider returns a model ID different from the selected one, label the workflow with the returned ID and do not report it as a 3.7 test.
What community reports can and cannot tell you
A Reddit discussion about Gemini 3.7 availability reports that the model disappeared from one selector; a separate Antigravity discussion compares 3.7 and 3.8 speed and quota experiences. These are individual reports from particular product surfaces, not evidence that Google retired the API model. Use Google’s current docs for support and check the exact client yourself. Availability discussion · Version experience discussion
Decide whether to migrate before browser testing
| Situation | Action | Evidence to keep |
|---|---|---|
| Client still offers 3.7 and reports that exact ID | Run a harmless read-only test | Provider response model field and timestamp |
| Old ID is redirected to 3.8 | Treat it as a 3.8 test | Actual response model and API route |
| Client cannot show the selected model | Do not compare model-specific results | Fix observability before evaluation |
| Agent cannot call local Tabbit CLI | Use a local supported client or manual browser work | Handoff note and expected page fields |
For browser setup, read the Tabbit CLI guide, existing-browser connection guide, and first-task checklist. The existing Gemini 3.7 model guide covers model details; a separate Gemini 3.8 model guide addresses the newer version.
More setup references
| Final check | What to inspect |
|---|---|
| Client entry | Confirm the actual skill command supported by your client |
| Local handoff | Confirm the same machine can launch Tabbit CLI |
| Page evidence | Keep a source URL and visible fields for each claim |
| Review point | Stop at a named, unsent draft |
Verdict
For a run on Gemini API, record 3.8 Flash as the model behind a request to the old 3.7 ID. Use a local client that can run Tabbit CLI to check the preview page, then compare its visible status with the named draft in Tabbit Browser.
FAQ
Can I still select Gemini 3.7 Flash?
Google AI Studio has removed 3.7 Flash from its selector. Google’s Gemini API documentation says requests to gemini-3.7-flash are automatically routed to 3.8 Flash.
Is Gemini 3.7 the same model as 3.8?
No. Google documents separate model IDs and release notes. Verify the response ID and do not treat results from one version as evidence about the other.
Can Gemini 3.7 use Tabbit Browser?
A compatible local agent can use Tabbit CLI to work with Tabbit Browser. Model selection, local CLI access, and website login are separate checks.
What should I do with an old Gemini 3.7 workflow?
Record the model ID, confirm whether the chosen client still offers it, update only if needed, and rerun a harmless read-only task before resuming work.
Can a remote Gemini agent reuse my local login?
Only if that service documents a local connection mechanism. Otherwise hand a task brief to a local agent or do the browser step yourself.