15 OCT 2026 — Anyone budgeting for Google's newest model should use the December price, not the launch price. Gemini 4 Argon, which Google calls its most capable model, became generally available on 13 October at US$2 per million input tokens and US$10 per million output. The API pricing page shows those are introductory rates: on 1 December they double to US$4 and US$20.
Build on the launch price and you have about seven weeks before the bill recalculates. This is the second time since August that Google has attached an expiry date to a Gemini launch price.
What the price list says
Google's Gemini API pricing page puts Argon at US$2.00 per million input tokens "through November 30, 2026" and US$4.00 "starting December 1, 2026". Output, which includes thinking tokens, moves from US$10.00 to US$20.00 on the same dates. Context caching doubles from US$0.10 to US$0.20. Argon is not available on the free tier.
The launch post by Koray Kavukcuoglu, Google DeepMind's chief AI architect, gives the US$2 and US$10 figures with a footnote marking them as introductory. It does not state the December rate in the body of the announcement.
A familiar pattern
Google did the same with Gemini 3.7 Flash in August, launching it at half its predecessor's price on a rate that expires on 31 December. Introductory pricing is a sensible way to court early users. It also means launch coverage quotes a number most customers will not be paying for most of the model's life.
From December, Argon's rate will match the US$4 and US$20 that TechCrunch reported for Meta's Muse Wonder when it launched a day earlier. The low launch price gives Google a cheaper headline than its new rival for now, while the long-run price puts the two level.
From vetted defenders to everyone
Argon first went to vetted security partners earlier this month through Google's Fairwind programme, with its cyber guardrails switched off, as we reported at the time. The general release keeps the guardrails on for everyone else. Google says it is now expanding Fairwind to "hundreds of additional" partners, and that defenders can apply to get access "without cyber guardrails on a rolling basis".
The launch post describes a "multi-layer defense stack" under Google's Frontier Safety Framework. Google says external biology red-teamers could unlock fewer than 1 per cent of the sub-tasks it tests for, and that harmful-request violations on an adversarial test set fell nearly two-fold from earlier Gemini models. Those are Google's own figures; the model card has the detail.
What Google says it can do
Google claims Argon ranks first on the Vals Index, a benchmark for finance, coding, legal and tax work, and that internal teams using its Antigravity coding tool increased agentic code submissions by more than 35 per cent while cutting rollbacks by nearly a third. The API accepts up to one million tokens of combined input and output, aimed at long-running agent tasks.
Budgeting for December
For cost estimates, model Argon at US$4 and US$20 from the start, not at the launch rate. Whatever you save before 1 December is a bonus. Our LLM cost calculator, a RECATOOLS tool, can run either set of prices.
View author profile → · Editorial policy
