Google releases Gemini 4 Argon, its new frontier model: Benchmarks, how to try it, and pricing

In the glow of a late‑afternoon screen, Sundar Pichai tapped his finger on a keyboard and posted a single line on X, announcing that Gemini 4 Argon would be live worldwide. The tweet, now vanished, marked the moment a new frontier model stepped out of Google’s labs and into the hands of developers and businesses. The launch follows a narrow window of testing with Google’s Fairwind Program, a cohort of trusted cyber defenders who had first handled Argon in September.

The journey began on September 30 when Koray Kavukcuoglu, Google DeepMind’s SVP, revealed Argon in a blog post. The model was initially shared only with a select group of cyber experts, a cautious rollout that mirrored Google’s history of staggered releases. When the public launch happened, it was not a surprise that the company kept the same name, but the scope widened dramatically: anyone with a Google AI Pro or Ultra subscription could now access Argon through the Gemini app on macOS.

Argon is not a casual chatbot. Google says it is engineered for “long, complex work” in software engineering, enterprise knowledge work like legal and finance, and cybersecurity defense. The model delivers fewer hedges, less repetition, and fewer clichés in everyday chat, while offering frequent progress updates on longer tasks. In practice, this means a developer can ask Argon to write a piece of code, and the model will provide incremental feedback rather than a single, opaque answer. For legal teams, the model can parse dense regulatory texts and surface key points without the usual fluff.

Google’s own evaluation shows Argon ahead of many rivals on several benchmarks. In the Artificial Analysis Intelligence Index dated October 13, Argon ties GPT‑6 Astra at 53 points with high reasoning enabled, one point above GPT‑6.1 Sol and the new Muse Wonder Max. The same index places Argon about five points behind Claude Opus 5.5 and Claude Sonnet 5.5. While the numbers are technical, they paint a picture of a model that is competitive in reasoning but still catching up in other areas.

For those who want to experiment, Google offers an introductory API price: $2 per million input tokens and $10 per million output tokens. Cached input tokens are discounted to $0.10 per million, a 95 percent reduction. The subscription tiers mirror Google’s broader strategy: Google AI Pro costs $19.99 a month, while the Ultra plan, introduced at I/O 2026, is now $200 a month after a previous $250 price. These figures place Argon’s cost below that of its main competitors, making it an attractive option for startups and enterprises alike.

As a writer who follows the pulse of the creator economy, I see Argon as a bridge between the raw power of frontier models and the practical needs of everyday professionals. It is not just another chatbot; it is a tool that can write code, analyze contracts, and defend against cyber threatsall in one interface. The move to make it widely available signals Google’s confidence that the model is ready for real‑world challenges.

Looking ahead, the real test will be how quickly developers adopt Argon for complex workflows. If the model lives up to its promise of frequent progress updates and reduced hedging, it could redefine how software teams iterate on code and how legal firms digest regulatory changes. The next few months will tell whether Argon becomes a staple in the toolbox of creators and enterprises or remains a niche product for power users.