Google announces Gemini 4 Argon AI model, but it is not public yet

In short: Google has announced a new AI model called Gemini 4 Argon, but most people cannot use it yet.

What happened

Google introduced Gemini 4 Argon, its next large AI model. The company says it does well at coding, office-style knowledge work, and cybersecurity. Even so, Google has not opened it up for general use.

Google says engineers inside the company are already using Argon. It claims the model helped save 300 TiB of memory across Google data centers, and it has been used to help move some software projects from C and C++ to Rust (a newer programming language that is often chosen because it can reduce certain types of bugs).

Google also shared test scores that compare AI models on set tasks, like giving students the same exam. On one software engineering test called DeepSWE v1.1, Google says Argon scored 77.9 percent, higher than several named rival models.

Google says Argon will support up to 1 million tokens of output. Tokens are chunks of text (think of them like word pieces), so a higher limit means the AI can produce longer answers in a single go.

Why it matters

Google is keeping the rollout slow because the first focus is cyberdefense. It is starting with a small group of trusted testers, including partners in its Fairwind Program. Security company Wiz is reportedly already using Argon and used it to find a serious flaw in a hospital-related system, although Google did not share details.

Google has not shared pricing or a timeline. It says broader access will start later with paid API users (companies that connect software to the model) and Google AI Ultra subscribers.

Source: Arstechnica