Gemini 4 Argon: What Google Has Confirmed, How Much It Costs, and Why You Should Read Benchmarks with Caution

Lead: Google has released its new flagship model, Gemini 4 Argon, for paying users. Find out who can use it, how much it costs, what the developers have confirmed so far, and what the limitations are.

Google announced Gemini 4 Argon on September 30, 2026, as a new “frontier” model for programming, working with enterprise knowledge, and cyber defense. For the first few weeks, it was available only to select defense personnel in the Fairwind program. It became generally available on October 13, 2026.

Who Can Use It

According to reports from October 13, the model is available via the paid Gemini API and Google AI Studio, in the Antigravity development environment for Ultra-level subscribers, and in the Gemini app as well as in AI search mode for Pro and Ultra plans. It is also available to Gemini Enterprise customers. Free accounts and the free API tier do not have access to it. Details may vary—for example, with Pro plans—so please check availability for your account directly with Google. We do not have information on availability in Slovakia.

Pricing

The introductory API price is $2 per million input tokens and $10 per million output tokens. Starting December 1, 2026, these prices will double to $4 and $20, respectively. The price for inputs from the buffer is $0.10 per million tokens. These prices are in dollars, exclude VAT, and do not apply to the Slovak market.

What the model offers

• Google sets a limit of 1 million outgoing tokens (previously 64,000).
• According to Google, this model is intended for software engineering, business intelligence, and cyber threat defense.
• The technical designation of the model in the API is gemini-4-argon.

What the benchmarks show

The data in the overviews comes from the manufacturer’s tables. In the DeepSWE v1.1 test, Argon scored 77.9%, while comparable models from other developers achieved approximately 74%. On the Vals index, it scored 68.9%. In the Terminal-Bench 4.0 test, however, it achieved 57.4% and lagged behind its top competitor (66.4%).

These figures are the manufacturers’ own values, not independent measurements. Only your own testing will determine how well the model performs on your code.

Limitations and Security

In terms of cybersecurity, the model performed well in defense-focused tests according to reviews, but offensive tasks remain blocked. According to one account, at launch, access was limited to a small group of defenders, and even then without standard security safeguards. This differs from the version available to the public.

Is it worth the cost?

We cannot honestly answer this question. Sources do not provide an independent comparison with other services, either in terms of quality or price for the average user. For developers, it makes sense to test the model on their own projects before November 30, while the lower price is still in effect. For regular users, the decision depends on the plan they already have and whether the service meets their needs.

What Has Not Been Confirmed

• That Argon will “beat” competing models. Performance data comes from the manufacturers.
• That it is a “direct response” to specific updates from another developer.
• Availability, prices, and plans for Slovakia.

Conclusion

The Gemini 4 Argon is a true new release with a clear release date, availability, and price. At the same time, it’s a model about which we currently know mainly what Google has said. The most sensible approach is to test it on your own tasks and compare the results and price after December 1.