The free tier gets one model
From Thursday 9 October, people using the Gemini app without a subscription will have access to exactly one model: Flash-Lite, the smallest in the family. Flash and Pro, which free users could reach before, go behind the paywall. Google has set out the new arrangement on its own support pages, which now list model availability tier by tier.
The tiers read as follows. Free accounts get Flash-Lite on what Google calls standard limits, refreshing every five hours. AI Plus, $4.99 a month in the United States, gets Flash-Lite and Flash at twice the standard limits — and loses Pro, which it previously included. AI Pro, $19.99 a month, gets Flash-Lite, Flash and Pro at four times the standard limits, plus Deep Think. AI Ultra adds the Ultra model and runs at five or twenty times the AI Pro limits depending on the subscription.
The changes apply to personal Google accounts. Free users see them from 9 October. Paid subscribers are being emailed individually with the date their own transition happens, which is an unusually staggered way to roll out a downgrade.
What a tier now means
There is a second control alongside the model list. Each available model can be run at a low, medium or high effort setting, and Google says the higher settings consume more of your limit. So the plan no longer buys access to a model so much as a quantity of model-time, divided by how hard you ask it to think.
That is a more honest description of what inference costs than a flat monthly fee implies, and it is also a quieter form of price increase. Nothing on the AI Plus invoice changes. What changes is what $4.99 reaches.
The move brings Gemini’s structure closer to the one OpenAI settled on, where the free product is a demonstration rather than a working tool and the useful models sit one payment away. Google had been the outlier among the large consumer assistants in letting free users touch a mid-tier model at all.
Who this lands on
Most people who use these assistants heavily already pay. The group affected is the one in between: users who reached for Flash occasionally, got a good answer, and never had a reason to subscribe. From Thursday they get a model built for speed and cost rather than for difficult questions, and the gap between what the free product can do and what the paid one can do widens by one full step.
The squeezed middle
The AI Plus change is the sharper one. A $4.99 tier that included the Pro model was the cheapest route to frontier-adjacent capability from a major vendor. Removing Pro from it leaves a four-fold price jump — $4.99 to $19.99 — between the model most people would want and the one they are now entitled to.
Google has not published a reason, and the support pages describe the state of affairs rather than justify it. The obvious pressure is the cost of serving Pro-class inference to a subscriber base counted in the hundreds of millions, at a price that was set when the free tier was a growth instrument rather than a line item.
What to watch is whether the effort settings become the real pricing lever. If a plan is defined by a budget of model-time rather than by a list of model names, the tiers can be retuned without any announcement at all, and a user would notice only that the same question now costs more of their allowance than it did last month.
