From 9 October 2026, according to an updated Google help article as relayed by Notebookcheck and 9to5Google, free Gemini users on personal accounts get one model, Flash-Lite, and the $4.99-a-month AI Plus tier loses access to Pro. Google has given no reason for the change. This piece is about the structure rather than the news: the new ladder leaves the cheapest paid rung with the least obvious value, and it shows that Gemini rations access with three separate controls that interact.

A note on sourcing. The tier table below comes from Google's own help article, but we have not read that article directly. We are relying on Notebookcheck and 9to5Google for the models and prices, and on two further write-ups (SmartScope and CreativeAINews) for the usage multipliers and context windows. Where those sources disagree, we say so.

What changes on 9 October

The free tier is the simple part. Per Notebookcheck, personal-account users without a plan lose Flash and Pro on 9 October and keep only Flash-Lite, which CreativeAINews describes as Google's smallest model. Notebookcheck cites Google's limits page for a 32,000-token context window on that tier. The change applies to the Gemini app on web and mobile, not to the Gemini API or Google AI Studio, according to CreativeAINews.

The paid tiers, as reported, look like this:

  • Free, $0: Flash-Lite only. Flash and Pro removed on 9 October.
  • AI Plus, $4.99 a month: Flash-Lite and Flash. Loses Pro. The help article gives no effective date; subscribers are to be emailed when the change reaches them (9to5Google).
  • AI Pro, $19.99 a month: Flash-Lite, Flash and Pro. Deep Think is reported as added at this tier by Notebookcheck, but see the conflict noted below.
  • AI Ultra, $99.99 and $199.99 a month: all three models. 9to5Google lists both prices without saying which maps to which Ultra plan.

The asymmetry is worth noticing. The free-tier cut has a date. The Plus cut does not. A person paying $4.99 for Pro access is losing something they bought, on a schedule they will learn only by email, while the free-tier user is losing something that was never promised on a schedule. Those are different kinds of change, even though the help article covers them together.

Three levers, one meter

The model list is only the first lever. Per Notebookcheck, Gemini has used compute-based limits since May: usage refreshes every five hours until a weekly cap is reached, the Pro model, Deep Research and media generation count for more than a simple prompt, and every available model carries a low, medium or high effort setting, where higher effort gives more thorough answers but consumes more of the limit.

It helps to separate what each lever controls. Model access sets the ceiling on what one request can cost: a plan without Pro cannot spend Pro-level compute on any single prompt, however it is used. The five-hour window controls the rate of spending, so a burst of heavy use is throttled and then replenished. The weekly cap controls the total, so a user who exhausts every window still hits a ceiling. The effort setting controls cost per request within a model, which means two prompts to the same model can draw down the meter at very different speeds.

Taken together, a 'message' is no longer a unit. The plan sells a budget of compute, and the user decides how to spend it by choosing a model, an effort level and a task type. That is a more honest description of the underlying economics than a message count, but it makes plans hard to compare, because Google describes limits relative to a 'standard' allowance rather than in tokens or prompts. We covered the OpenAI version of the effort lever in our earlier piece on the GPT-5.6 reasoning slider; the Gemini version differs in that effort draws down a shared meter rather than selecting a separate price.

Pricing the steps

SmartScope and CreativeAINews both report the same multipliers: free is 1x standard usage, AI Plus 2x, AI Pro 4x, and AI Ultra 5x or 20x the AI Pro limit. CreativeAINews adds context windows of 32,000 tokens (free), 128,000 (Plus) and 1 million (Pro and Ultra). These come from secondary sources, so treat the arithmetic below as ours, not Google's.

Start with the price steps. Going from free to Plus costs $4.99 and adds one model (Flash). Going from Plus to Pro costs $15.00 more ($19.99 minus $4.99) and adds Pro. That second step is 3.0 times the first ($15.00 divided by $4.99 is 3.006). Put differently, the rung that now has the least to offer is also the cheapest, and the rung with the model most people actually want is three times as far up the ladder.

Now divide by the multipliers. Plus delivers 2x standard usage for $4.99, or about $2.50 per standard unit. Pro delivers 4x for $19.99, or about $5.00 per unit. The marginal step from Plus to Pro buys two extra units for $15.00, or $7.50 each. If the Ultra multipliers are read at face value (5x Pro is 20x standard, and 20x Pro is 80x standard) and if the $99.99 plan is the smaller one, which no source confirms, Ultra works out to about $5.00 per unit at the lower price and $2.50 per unit at $199.99.

Two cautions apply. A 'unit' of standard usage is not a fixed amount of compute, since Pro and effort settings consume more of it, so a unit on Plus (Flash and Flash-Lite only) is not the same purchase as a unit on Pro. And the mapping of Ultra prices to multipliers is our assumption. What the figures do show is that Plus is the cheapest way to buy metered usage and the most expensive way, per dollar, to learn that Pro is not included. The price differences are pricing the model, not the usage.

What does Google expect from this? It has not said, so any reading is interpretation. One reading is that Pro becomes the conversion point: users who relied on it through Plus or the free tier now face a single $19.99 decision. Another is that Plus is repositioned as 'more Flash' for people whose needs are modest. Both are consistent with the numbers. Neither is stated by Google.

The counter-argument: free compute is not a right

The strongest defence of the change is economic. A free tier that includes a mid-sized model and a large one is a subsidy, and a lab running a compute-based meter has an obvious incentive to move the expensive model behind payment. CreativeAINews notes that The Decoder interpreted the change as setting up a pricier model, but that is The Decoder's reading, not something Google has stated. On this view, narrowing the free tier to the smallest model is a predictable correction, and the five-hour and weekly limits were already evidence that Google was rationing.

The weaker part of that defence is Plus. Tightening a free tier removes a courtesy. Removing Pro from a paid plan removes a product feature, and the help article leaves the date open. A fair reading of the economics can accept the first change and still question the second.

What we can't tell yet

Several details are unresolved in the coverage we reviewed:

  • Google's own pages disagree. Per Notebookcheck, the US plans page still lists free-tier access to 3.6 Flash and varying access to 3.1 Pro, and the general limits article still shows all three models on every plan. Only the new help article is dated.
  • Deep Think is contested. Notebookcheck says Pro gains it but notes that Google's limits page lists it as Ultra only. 9to5Google describes it as currently available only on the $99.99 and $199.99 plans. CreativeAINews says it was previously Ultra only and is now on Pro and Ultra.
  • The Plus effective date is unannounced, and the free tier's Flash-Lite version is not stated by Google (9to5Google reports it as 3.5 Flash-Lite).
  • The article reportedly covers personal accounts and says nothing about work or school accounts, or about whether existing chats switch models.
  • Ultra price-to-multiplier mapping, which our per-unit figures depend on, is not confirmed.