Free Gemini drops to Flash-Lite on October 9. Rerun your work on it before the switch

Plenty of small firms run part of their work in the free Gemini app: a summary here, a first draft there, a weekly report nobody formally assigned to a model. On October 9, the model under that work changes. Picking which work to hand to AI is covered in our worksheet on what to hand to AI first. This post covers what to do when a vendor changes the model under work you have already handed over.

What changed

On October 3, 9to5Google reported an updated Google help page on Gemini model access. Per that report:

Free users. Starting October 9, people who do not pay for Gemini lose Gemini 3.6 Flash and Gemini 3.1 Pro. They are limited to Gemini 3.5 Flash-Lite.
AI Plus ($4.99 a month). Keeps Flash-Lite and Flash and loses Pro. These subscribers will get an email that explains when the change takes effect for them.
AI Pro ($19.99 a month). Keeps all three models and gains Deep Think, which the article describes as “maximum parallel reasoning.”
AI Ultra ($99.99 and $199.99 a month). No regression on this model grid. Deep Think was already there.

The model grid comes from 9to5Google’s reading of Google’s page. The prices are the ones 9to5Google states for those plans. The article is about the Gemini app. It does not settle whether Workspace accounts or the Gemini API are affected, so check those separately.

Why this matters more to a lean firm

A large company notices a model change because someone owns the vendor contract. In a business of one, the free app often became a process quietly. One person used it every Friday, and now a client deliverable depends on it. Nobody wrote down which model it ran on, because the app picked the model for them.

That is the exposure. Losing Pro costs nothing in dollars. It costs you if a process that worked last week starts producing weaker output and nobody notices until a client does.

The four-day test

You have until October 9. This test needs no new tools.

List what runs on free Gemini. Ask everyone, including yourself: what did you use the Gemini app for in the last two weeks that ended up in front of a client or in a decision? Write each one as a single line: the process, who runs it, how often.
Pull five real inputs per process. Use last week’s actual material: the five emails you summarized, the five briefs you drafted. Keep the output you shipped at the time as your reference.
Rerun all five on Flash-Lite. If the app’s model menu lets you pick Flash-Lite now, do that, so you test the model you will be left with rather than the one you have. If it does not, run this step on October 9 instead.
Score each run pass or fail against one written check. Pick the check before you look at the output. For a summary: are all action items and dates present and correct? For a draft: could you send it after light edits, yes or no?
Apply the rule below.

The decision rule

For each process, after five runs:

5 of 5 pass: stay on free. Write “Flash-Lite” and the test date next to the process so the next change is easy to spot.
3 or 4 pass, and the failures are easy to catch: stay on free, but add a named person who checks every output before it leaves. Easy to catch means the person can see the miss in under a minute.
2 or fewer pass, or the failures are hard to catch: that process needs more than Flash-Lite. Rerun the same five on Flash. If Flash passes, the $4.99 plan covers it, but watch for Google’s email with your own date.
Only Pro passes: your choices are the $19.99 plan, a comparable model from another provider, or taking the process back by hand. Price each one against what the process is worth to you per month.

The test works best on work that is easy to check, which is the same reason that work was a good first handoff. If you cannot write a pass-or-fail check for a process, that is worth knowing too. It means you were trusting the output without a way to judge it, whatever model ran it.

A small worked example

Say a two-person design studio runs three things in the free Gemini app: meeting notes into action items, first drafts of project proposals, and alt text for client image libraries.

They run the test. Meeting notes pass 5 of 5 on Flash-Lite, so they stay free. Alt text passes 4 of 5, and the miss (a generic description) is obvious at a glance, so it stays free with the designer checking each batch. Proposals pass 2 of 5 on Flash-Lite and 5 of 5 on Flash, so the owner moves that one process to the $4.99 plan. Total new cost: $4.99 a month, plus one page that says which model each process runs on.

That example is illustrative, not a client result.

What to do this week

Today: list every process that touches the free Gemini app, with an owner for each.
By Wednesday: run five real inputs per process on Flash-Lite, scored against one written check.
By Thursday: apply the decision rule and change plans for any process that needs Flash or Pro.
Before October 9: write the model name and test date next to each process, and check your Workspace and API usage separately.
After October 9: rerun one input per process to confirm nothing moved.

Model access can change again, from Google and from everyone else. A list of processes, each with its model, its check and its owner, turns the next change into a one-hour job. If you would rather build that list with someone, a consult block is a reasonable place to start.