Gemini Pro Drops Out of AI Plus: The Model Tier Is the New Price Tag

The fact: on 9 October the plan's contents change

On 9 October Google rewrites what its Gemini plans contain. Customers paying 4.99 euros a month for AI Plus keep Flash-Lite and Flash, and watch the Pro model disappear.

Anyone opening the Gemini app at zero cost drops to the Flash-Lite models. The change travels on a support document, as reported by heise^[1], which also recalls the usage-limit revision already made in May.

Keeping the Pro model requires the AI Pro plan, 21.99 euros a month. Above it sits AI Ultra, from 99.99 euros a month. The two top plans add Deep Think, the company's heavy-duty parallel reasoning.

Google is emailing the AI Plus customers affected by the change, with the date that applies to each of them.

What this move really is

Here the list price stays identical.

What changes is the model tier that price buys, and the change lands on a help page, with zero press releases. This is the cleanest form of a price increase: same receipt, lighter product.

The enterprise AI market knows this lever well on the API side, where the model tier governs cost per million tokens. Now the same lever appears on the subscription side, where the customer's bargaining position is weak. The AI model tier becomes a pricing variable in its own right, alongside price and volume.

A vendor moves it in an afternoon, and the customer notices the difference in answer quality before seeing it on the invoice.

The competitive shift: from price to tier

For years the comparison between vendors ran on two numbers: price per token and public benchmark scores. That grid is losing weight.

The new axis is the tier guaranteed by contract.

Two offers at the same price are worth different amounts when one promises the top tier and the other leaves the vendor free to scale downward. Anyone buying generic "AI capacity" is buying a good that deflates in price and shifts in substance. The vendor holding the right to change tier at its own discretion defends a sturdier margin than the one showing the highest score.

The move from price to tier also changes the buyer's job, which now means measuring output on a basket of real cases. The market has moved.

Who feels it

The first group covers teams that built workflows on top of a consumer plan. Prompts tuned for the top tier deliver less on a lightweight model, and the tuning work comes back on the table.

The second group involves departments that booked the subscription as a fixed cost per seat. That cost stays flat on the spreadsheet, and falls in output. The expense line lies to the CFO.

The third group sells downstream services.

A consultant promising a given output quality promises it on a model tier governed by someone else. It holds for Google today, and it holds for any vendor tomorrow. The useful clause names the model, states the tier and sets a notice window.

The market signal

The market signal: the model tier is the margin lever of AI vendors.

The price of the frontier tier falls with every release, and vendors recover by moving customers between tiers. The public price list stays a shop window, and the substance lives in the support notes.

Procurement teams read those notes the way they read software maintenance contracts fifteen years ago. The useful question changes shape: instead of "what does it cost", it becomes "which tier do you guarantee me, for how many months, with how much written notice".

The same holds for internal consumption, where a silent downgrade pushes human working hours downstream. A team that rewrites its prompts after every tier change pays the bill in days, and those days cost more than the subscription difference.

The model tier enters the contract

Compliance pushes in the same direction.

The European market for AI Act tooling is drawing players such as Collibra and Trail ML, as The Next Web^[2] reports. Documenting which model runs inside a business process becomes a requirement with legal weight.

A silent tier substitution cracks that documentation. An AI systems register declaring "Gemini Pro" loses its value the moment the plan serves Flash. The audit trail describes a product that has changed underneath.

Control over the AI tier stays in the vendor's hands on devices too: The Verge^[3] covered a tool that removes Apple Intelligence from a Mac, a sign of the demand for local control over features that arrive from above. Anyone signing on behalf of a company asks for the same thing, put in writing.

The board's four questions

For the head of strategy the urgent question is single-vendor dependency in the products going to market next quarter.

For the CFO the line to revisit is AI subscription spend per seat. A 4.99-euro plan that loses the top tier pushes toward 21.99 euros, meaning a factor of four on the same user base. A simple calculation across a thousand seats gives more than 200,000 euros of difference a year.

For the head of digital the vendor portfolio needs rereading by tier served, plan by plan.

For investors, the thesis confirmed is the model as a commodity: value moves to deployment and compute, where contracts last years. The valuations of compute capacity providers tell the same story better than model scores do.

What to decide in the next 90 days

Three moves fit inside a quarter.

• Inventory: list every business workflow running on a consumer plan, with the name of the expected model.
• Clause: ask the vendor for the guaranteed tier and 90 days of written notice on any downgrade.
• Test: keep a second vendor warm on two use cases, and measure output every quarter.

The next step concerns the budget. Anyone who put AI in the books as a cost per seat should move it to a line with two parameters: model tier and volume. A change like the one on 9 October then enters the numbers within an hour, instead of surfacing at year end.

The lesson stays simple: in 2026 the vendor sells a level of intelligence, and that level lives in a table the vendor updates on its own. Whoever writes the contract writes the product.