Developers get seven weeks of Argon at introductory pricing before it rises to $4
Google has begun the broad rollout of Gemini 4 Argon, but free users and lower subscription tiers are left out. Developers get an introductory price of $2 per million input tokens — but only until November 30.
Argon, Google's newest frontier model, was announced on September 30, but at that point it was available only to a limited circle. Now it is reaching the broad public — provided you pay. Both Ars Technica and The Verge reported on October 13 that the rollout has begun, but the access map is tiered, the prices are time-limited, and Google has not said when or whether non-paying users will get access.
Who can use Argon today
According to Ars Technica, Argon is coming only to paying users "for now," and even lower subscription tiers do not get full access. The map looks like this:
- The Gemini app: AI Pro and AI Ultra subscribers get the model. Pro subscribers, however, are limited to the app's Canvas mode, according to Ars Technica.
- Antigravity and AI Studio: Here Argon is reserved for AI Ultra subscribers. Ars Technica writes that "in Antigravity and AI Studio, the new model will only be available to those with AI Ultra subscriptions."
- Search: Here the sources diverge. The Verge reports that Argon is rolling out to Pro and Ultra subscribers in the Gemini app "and AI Mode in Search," in addition to Ultra users on Antigravity. Ars Technica does not mention Search at all. The disagreement cannot be resolved from the available information.
For those left out: Gemini 3.1 Pro remains available, Ars Technica reports. And to the question lower tiers are asking — when does Argon come to Plus users, free users, or lower Antigravity tiers? — Google has, according to Ars Technica, "declined to clarify when (or if) Argon will come to Antigravity for other subscription tiers." The Verge also describes access for Plus subscribers and free users as unclear.
The developer price with an expiration date
The most concrete element in the story is the API pricing, which is aggressive — but temporary. Ars Technica sums up the terms like this: "Pricing is temporarily set at $2 per million input tokens and $10 per million output tokens, with a 95 percent discount on cached tokens. This deal is only good through November 30, at which time the rate will jump to $4 per 1M input tokens and $20 per 1M output tokens."
That gives developers a concrete deadline: anyone planning projects around Argon has roughly seven weeks to build at the introductory price before it doubles. The 95 percent discount on cached tokens is in itself a significant detail for workloads where much of the context is repeated between calls.
The Verge puts the price in context: the introductory price corresponds to roughly a fifth of the cost of OpenAI's GPT-6 Astra and Anthropic's Fable 5 models. The outlet writes that even after the price doubles at the end of November, Argon will "still be significantly cheaper than its frontier rivals on a per-token basis." Both the level and the planned doubling point toward Google positioning the model below its competitors on price — that is an analysis of the pricing structure, not a strategy Google itself has described in the sources.
The road here: first cyber defenders, then API customers
The rollout has proceeded in several stages. According to The AI Insider, Argon was at the September 30 announcement "initially available only to a select group of cyber partners through Google's Fairwind Program," the company's security initiative. Google says the model is trained for defensive cyber work.
Then, according to eWeek, building on Google's rollout plan, "paid API customers and Google AI Ultra subscribers are next in the rollout, followed by additional developers and businesses using AI in enterprise applications." Ars Technica notes that Google itself emphasized the need to "move cautiously toward a full release because of the potential for nefarious use."
What Google itself claims — and what employees say
Google's performance claims are sweeping, but they are the company's own figures, not independently verified. According to a comparison table Google has published, Argon was the sole leader in 13 of 19 benchmark rows and tied leader in one row — that is the reading of analyst 池田朋弘, reporting from Google's table. eWeek counted the same table at 14 of 19 rows. Both are interpretations of Google's own material, and the discrepancy between them should be noted.
On the capacity side, Google (relayed by 池田朋弘) points to an output limit of one million tokens, expanded from 64,000.
The internal examples Google highlights are striking if they hold up: Argon is said, according to the company, to have identified redundant emoji rendering in YouTube's comment system, which when removed lowered server load by 45 percent. Engineers are also said to have turned the model loose on Project Wing's proprietary drone-physics simulator, where it found critical bugs that had been in the codebase since 2018. These are company claims reported by Ars Technica — not independently verified.
A more nuanced picture comes from Business Insider, which, according to unnamed employees and internal documents, reports that Google tested the models Argon, Barium and Carbon internally, and that the internal "Barium-B" became the public Argon. The reception was mixed: one employee reportedly said early versions of Argon felt behind on some coding tasks, comparing the model to Claude Opus 5, an older Anthropic release. That is worth keeping in mind when reading the benchmark leads: Google's own numbers and parts of the internal impressions do not point unambiguously in the same direction.
What remains open
Several central questions remain unanswered:
- Free users and Plus subscribers have no confirmed path to Argon. Google has, according to Ars Technica, declined to clarify when or whether the model will reach lower tiers, so no timeline can be given.
- The rollout scope is unresolved between the two main sources: The Verge includes AI Mode in Search, Ars Technica does not.
- The benchmark figures come from Google's own comparison table and have not been independently verified — this applies both to the 13-versus-14 count and the internal efficiency examples.
- The duration of the paywall is unknown. That Argon is reaching only paying users in this rollout is confirmed; no source says how long that restriction will last.
What is confirmed as of today is the economics and the deadline: paying Pro and Ultra customers and developers can start using Argon at $2 and $10 per million tokens — but only until November 30. After that, $4 and $20 apply, still, according to The Verge, below the prices of GPT-6 Astra and Claude Fable 5. How widely available the model becomes over time, and how long the paywall stands, remain for Google to clarify.