Google spent early October selling Gemini 4 Argon as the cheapest route to frontier AI. By mid-month the same company was emailing paying customers to tell them the cheap route is closing.
Two deadlines landed within 48 hours of each other, and both point in the same direction.
On November 13, 2026, Google AI Plus loses the ability to select Pro models in the Gemini app and in AI Mode inside Google Search.
The notice that went out to subscribers is unusually direct about what survives.
"Starting on 13 November 2026, the ability to select Pro models in both the Gemini app and AI Mode in Google Search will only be available to Google AI Pro subscribers and above. You’ll still have access to our latest Flash-Lite model and Flash, built for speed and intelligence."
Existing chats that were built on a Pro model can be continued, but only with a different model. That is the entire consolation prize.
Want to keep Pro-level model selection? The subscription price moves from $4.99 to $19.99 per month.
Four times the money, roughly two weeks after Argon was framed as a broad capability upgrade for the whole Gemini family.
The Price Advantage Was Always a Promotion
Developer pricing carries its own expiry stamp.
Through November 30, 2026, Gemini 4 Argon costs $2 per million input tokens and $10 per million output tokens.
From December 1, those figures become $4 and $20.
Both rates double. So does cached input, which moves from $0.10 to $0.20 per million even after a 95% cache discount.
Google’s own API documentation uses the word introductory. Nothing here is hidden, and that is exactly what makes it worth reading carefully.
- Gemini 4 Argon, introductory rate: $2 input / $10 output per 1M tokens
- Gemini 4 Argon, standard rate from December 1: $4 input / $20 output
- Claude Opus 5.5 as listed by Artificial Analysis: $4 input / $20 output
- OpenAI GPT-6 Astra: $10 input / $50 output
At the introductory rate Argon undercuts the field. At the standard rate it simply matches Anthropic’s listed Opus 5.5 price and stays cheaper only than Astra.
That is not a pricing revolution. It is a launch discount with a calendar attached.
The mismatch matters because launch coverage reads the $2 line, while billing systems read the $4 line.
Benchmark-driven cost estimates make the cliff sharper than the raw token table suggests.
Artificial Analysis estimates Gemini 4 Argon (High) at $1.99 per task on its Intelligence Index workload.
GPT-6 Astra (Max) lands at $3.26 per task.
After the December increase, the same Argon estimate rises to roughly $3.98.
A gap that looked like a 40 percent advantage turns into a rounding error, and only if token usage stays flat.
Where the Benchmarks Actually Land
The performance story is as narrow as the price story.
Artificial Analysis scores both Argon and Astra at 53 on the Intelligence Index in its high-effort versus maximum-effort comparison.
Claude Opus 5.5 scores 58 at maximum effort, 54 at high effort, and 51 at medium.
The ordering shifts with the setting, which is a polite way of saying nobody owns a decisive lead.
Google DeepMind positions Argon for complex software engineering, enterprise knowledge work, and cybersecurity defense.
That target list is serious, and it is also the exact category of work where token counts balloon.
Deep dives, long agent runs, and multi-hour refactors do not behave like chat turns.
Which is why cost per completed task is the only number worth arguing about. Caching helps the predictable case.
A 95% cache discount rewards repetition, stable context, and high-volume prompt reuse.
It does far less for the one-off, long-horizon agent run that generates the alarming invoice.
To be fair, none of this is unusual behavior. Introductory pricing, tier reshuffles, and paywall tightening are standard moves across the industry.
The unusual part is the compression. A flagship model launch and a paid-tier downgrade were announced inside the same month.
That speed is what should make buyers cautious, not the price increase on its own.
Cheap tokens are a customer acquisition cost, and acquisition costs eventually end.
Every aggressive launch price is a bet that switching later will cost more than staying.
Gemini 4 Argon is a strong entry in that playbook rather than an exception to it.
Before November 13, three checks are worth doing.
- Confirm which models your plan can actually select after the cutoff date, in the app and in Search, not just in the marketing page.
- Do the math on the jump from $4.99 to $19.99 against how often you genuinely reach for a Pro model rather than a Flash model.
- Check what else is bundled into the plan, since storage, Workspace perks, and video subscriptions often justify the tier better than model names do.
For developers, the exercise looks different.
- Measure cost per completed task, not cost per token, before December 1.
- Model the standard rate into any roadmap that depends on cheap long-context work.
- Keep a second provider wired in, because a doubling rate is a real migration trigger.
There is one more detail worth holding onto. Recent coverage notes that Argon ties for third place on a benchmark Google itself helped build.
That is not an accusation, it is a reminder to read leaderboards next to the invoice.
None of this makes Gemini 4 Argon a bad model.
Its caching economics, long-context behavior, and agentic tooling are genuinely competitive, and for many teams it will still be the right default.
The problem is expectation management. The marketing runs ahead of the billing system.
Both sets of numbers are real. Only one of them arrives in December.
So the honest summary of Google’s Gemini month is a company shipping its most capable model yet while quietly charging more to keep using it.
That is not a scandal. It is a strategy, and it is worth naming as one.
Comments