· AI
Google Already Told You Gemini 4 Argon's Price Would Double. It Put the Date in a Footnote.
Nobody hid this price increase. Almost nobody will notice it anyway.
Google's own pricing documentation for Gemini 4 Argon lists two columns: $2.00 per million input tokens and $10.00 per million output tokens through November 30, 2026, stepping up to $4.00 and $20.00 starting December 1, according to the current Gemini API pricing page. Google said as much back in September, when it introduced Argon at what it called an introductory price and added a footnote warning that once that period ends, the $4 and $20 rates take over, per Google's own launch post. That post did not name a date. The pricing page now does. Every rate on the page, input, output, cached input, batch processing, doubles on the same morning, and Google's own flagship Gemini 4 Carbon gets the identical treatment.
My position: this is not a gotcha story about Google hiding something. The date and the numbers are both published, in writing, by Google itself. The story is that disclosure and prevention are two different things, and a lot of companies are about to learn that distinction with their December invoice.
The steelman: Google did this the honest way
A launch discount that converts to a real price on a named date is a more honest version of SaaS pricing than most vendors run. Plenty of software raises prices with a quiet renewal email after a customer is already locked into a workflow. Google put the after-price in the same announcement as the before-price, with a specific date now attached to both. A buyer who reads the page they signed up against has everything needed to budget correctly. Judged purely on disclosure, there is nothing to complain about here, and I would not want a regulator banning introductory pricing just because some buyers never read the page twice. That would raise the cost of launching anything new, including the next cheap open-weight competitor this blog would normally be rooting for.
Where the steelman runs out
Disclosure is not the same thing as a plan getting executed. One cost-tracking analysis of the change notes that the jump is easy to miss precisely because nothing on the launch page flags it for someone who priced out a project in October and moved on. A second analysis of the same pricing table draws a sharper contrast: Argon's launch rate buys a lower price on a model that spends more tokens to do each job, while OpenAI and Anthropic's recent cuts came from newer models doing the same work in fewer tokens, not just charging less per token. That matters here because Artificial Analysis's benchmark run found Argon generated 113 million output tokens across its evaluation suite, well above the 65.1 million median for the models it tested against. A verbose model is the worst candidate for a doubling output price, since output is exactly the number that moves the most tokens through the bill.
The math a lot of teams skipped in September
Take a workload running 10 million input tokens and 50 million output tokens a month, a reasonable size for a document or support pipeline. At the launch price, that is $20 on input and $500 on output: $520 a month. On December 1, with zero change in traffic, the same workload costs $40 on input and $1,000 on output: $1,040. Nothing about the product changed overnight. The calendar did. If a team sized its budget, or worse, a client proposal, off the number on the page in October, that plan is already wrong, and it was wrong the day it was written, not the day the price changed.
Google bundled two more changes into the same date that are easy to miss for a different reason: no dollar sign attached to notice. Gemini 2.5 Pro and Gemini 2.5 Flash are deprecated as of December 1, and requests to either model get rerouted automatically to newer models billed at the newer rates, per the same pricing documentation. An app built on the older models, with nobody planning to touch the Argon pricing page at all, still gets a new bill that same morning.
What this actually means for a business running on any of this
The lesson is not to avoid introductory pricing or to distrust Google. It is that a price schedule with a named future date is a calendar entry, not a rate you lock in by reading it once. Put the date on the calendar the day you commit to a model, not the day the invoice arrives. Re-run the math at the standard rate before you ship anything you intend to run past that date, and if the economics only work at the launch price, you have not actually found a cheap model. You have found a cheap month.
This is the exact gap MojoAI checks before signing off on a client's model choice: not just the per-token rate today, but what it becomes on the date the vendor already told you about. If your AI budget runs on a number with an expiration date nobody wrote down, that is worth a look before the invoice does it for you.
Sources
References used in this article. Links also appear alongside the relevant claims.
Let's make it happen.
You don't need a polished brief. A couple of lines about where you want your company to go is plenty, and we'll come back with what we'd tackle first.
We scope the work around your goals and timeline before quoting anything, so you know exactly what you're getting.