Google's Model Rotation Is Getting Faster—And That Breaks Your Testing

2026-09-03 — customer acquisition strategy India

Google pushed Gemini 3.8 Flash into AI Mode on release day. Paid subscribers with Pro and Ultra accounts can pick it from the "+" icon, which is how you'd expect a model launch to work. Except the pace they're moving at now makes that statement almost useless for anyone actually testing performance.

This is the third Flash release in six weeks. Actually, let me be more precise—it's the third in six weeks in Search specifically. That's not a typo. That's the new normal at Google.

If you tested AI Mode citations on 3.7 Flash in mid-August, you were working with a model that doesn't appear in the dropdown anymore. It's gone. Replaced. The screenshot Google shared today shows 3.8 between "Auto" and "Pro," exactly where 3.7 sat two weeks ago. But 3.7 isn't visible now. You can't select it.

This matters more than it sounds.

When you're comparing results week to week or even day to day, you need to log which model you're testing. Not just note "we tested AI Mode" but actually write down "3.8 Flash" or "3.7 Flash." Because the changes between them aren't always improvements you can see on your own content. Sometimes they're invisible shifts in how the model reasons through a query, and a gap between one test and the next could mean Google rewired something fundamental, not something your site did wrong.

The free-tier users? They're still stuck with whatever the default is. That hasn't changed, at least not yet. But paid users keep getting new options pushed to their menu every few weeks, and Google's support page hasn't updated since 3.7 landed. So the only proof that a model even answered your question is your own screenshot.

What happens next is worth watching. Both earlier Flash models—3.5 and 3.7—eventually became the default. That's the path forward. If 3.8 follows the same route, it will eventually roll out to free users too. But until that happens, there's no official documentation saying which model is doing the work. You're flying blind without your own notes.

Most consultants get this wrong, including us sometimes. They test a paid feature, see movement in citations or ranking, and don't realize they're comparing data from two completely different models. It looks like progress. Might just be noise.

Source: "Google just released a new Gemini 3.8 Flash model for Google AI Pro and Ultra subscribers today, three weeks after the release of Gemini 3.7 Flash." — Thurrott.com