Google's Gemini 4 Argon reached paying developers and consumers on October 13, 13 days after the company first announced it on September 30. The introductory price is $2 per million input tokens and $10 per million output tokens, and it doubles for anyone who starts after November 30. That deadline gives budget-minded teams a reason to test quickly and a reason to plan carefully.
Google's launch post says Argon is available to developers and enterprises through the API in Google AI Studio and the Gemini platform. Google AI Ultra subscribers can use it in Google Antigravity, and Ultra and Pro subscribers can use it in the Gemini app and AI Mode in Search. The release ends a closed period in which Argon was limited to vetted cyber defenders and a select group of partners in Google's Fairwind Program.
Gemini 4 Argon price and the November 30 step-up
The introductory rate is aggressive. Cached input tokens cost 95 percent less than standard input, which matters for workflows that resend the same long documents or system prompts. After the promotion, the standard price becomes $4 per million input tokens and $20 per million output tokens, starting after November 30, according to Forbes and AlphaCorp.
Forbes reads the step-up as deliberate. It frames the November 30 deadline as a bid to lock developers into Google's stack before prices rise. The only confirmation so far is a Google footnote, as Forbes reported it, so teams should check the official pricing page for any change.
A simple example shows the effect. A workload of 10 million input tokens and 2 million output tokens per month costs $40 at the introductory rate and $80 at the standard rate. That math uses only the published per-token prices, without any usage tiers or negotiated discounts, but it shows how quickly the step-up doubles a line item.
Who can use Gemini 4 Argon right now
Paying customers have the broadest access. Developers and enterprises can call Argon through the paid Gemini API, which matters most for teams building coding assistants or tools that process long documents. Consumers get Argon in the Gemini app only on paid Google AI Pro and Ultra tiers.
The free plan does not include Argon, and neither does Google AI Plus, according to FelloAI's rollup. Google has not said whether or when those users will get access. Readers who want Argon without a subscription will have to wait.
The cyber-focused history also leaves a question open. Forbes reported that partners received a version without cyber guardrails. It is unclear whether that build is the same product the public now gets, which matters for anyone planning to use Argon for security work.
How Gemini 4 Argon benchmarks compare with GPT-6 Astra and Anthropic's models
Google has pitched Argon as a workhorse for coding and cybersecurity. TechCrunch reported that Google claims Argon scored significantly higher than OpenAI's GPT-6 Astra and Anthropic's Fable and Opus models across a range of benchmarks. Those are Google's own numbers. Shattered quotes a Google description of the model as the "next era of frontier intelligence," but that is company marketing rather than an independent measurement.
The rest of the evidence is also mostly self-reported. AlphaCorp says Argon led 13 of 19 benchmark rows in Google's comparison table, citing a third-party blog. The brief for this story did not verify that count against Google's table. Box CEO Aaron Levie reported that Argon scored 14 percentage points higher than Gemini 3.1 Pro across Box's tests, according to Forbes. That is one company's internal evaluation.
AI Weekly reports that Google SVP Koray Kavukcuoglu said Argon ranks first on the Vals Index for coding, finance and legal work. AI Weekly describes this as a paraphrase in an aggregator summary, not a verbatim quote, so treat it as reported rather than confirmed.
The practical read is simple. Argon may well be strong at coding and document work, but the public comparisons available today come from the companies that built or tested the models. Independent results will matter more than launch tables.
Which token limits to trust
The token specifications also need checking. Shattered describes a shared 1,048,576-token window, which it compares with a prior 64,000-token output ceiling. AlphaCorp describes a 1-million-token output limit. Those are different specifications, and the sources do not agree on which one the million-token figure describes.
Developers should check the model documentation in Google AI Studio before designing around any limit. A context window and an output cap affect a product in different ways, and getting that wrong can break a long-document pipeline in production.
The White House accord and the rules around Argon
Argon's launch came one day after a voluntary White House AI accord signed September 29. Technology.org reports that signers described the accord as "morally binding." The accord creates no penalties and no regulator, so compliance depends on the companies themselves.
Google's participation in the U.S. government's voluntary pre-release review follows the same logic. Forbes reported that Google said it was taking part, but the process carries no enforcement mechanism. The formal route is different. USA 24 has reported on the FTC's civil investigative demands to OpenAI and Anthropic, which carry legal force that a voluntary accord does not.
How to test Gemini 4 Argon before November 30
Teams that want to evaluate Argon can follow a few steps that keep costs and claims in check.
- Check your provider's current pricing page before November 30, since the standard rate starts then.
- Build a test set from your own tasks, such as code review tickets or contract clauses, and score outputs against answers your team has already verified.
- Estimate monthly spend at both rates using your real input and output token counts.
- Use cached input for repeated long prompts, because the cached rate is 95 percent below standard.
- Treat vendor benchmark tables as claims until your own results match them.
Platform defaults can also change under you. Replit recently said OpenAI replaced Anthropic as the default model for Replit Agent, which is a reminder that the model behind a tool you rely on can shift without much notice.
What comes next for the Gemini 4 family
Argon is the only Gemini 4 model available so far. Google has said only that it looks forward to bringing the rest of the Gemini 4 family to users, and it has not given dates. Readers should watch three things: whether the November 30 price increase takes effect as Forbes reported, whether free and Google AI Plus users gain access, and whether independent tests reproduce Google's benchmark lead. The answers will show whether Argon's low introductory price was a launch offer or a lasting position.
