Google has unveiled Gemini 4 Argon, a new artificial intelligence model that the company calls the most powerful in its lineup. Argon is designed for complex tasks in programming, engineering, research, and text work, but Google sees cybersecurity as its main area of application.

For now, the model is not available to everyone: the company has provided it to a limited number of security partners as part of its Fairwind program. Argon was specifically trained for defensive cybersecurity tasks, and Google claims that the model can autonomously find, verify, and fix critical vulnerabilities in software.

AI that finds vulnerabilities and fixes them itself

Google is betting on Argon’s ability to carry out long sequences of related actions without constant human involvement. The company says the model was built to handle complex tasks that require deep reasoning and sustained execution.

In cybersecurity, this means a shift from simply generating recommendations to working with code more independently. Argon is expected to detect a potential vulnerability, verify whether it actually exists, and then prepare and apply a fix.

That is why Google is not yet opening the model for broad use. Argon is being distributed among selected security partners through Fairwind, the company’s program dedicated to protecting digital infrastructure.

The model is already being used inside Google

Argon’s capabilities are not limited to cybersecurity. Google also positions it as a strong model for programming and engineering tasks.

Company employees are already using Argon in their day-to-day work — in particular, to find and fix bugs in code and to migrate existing codebases.

Another claimed capability of the model is related to analyzing visual information. Argon can work not only with text and code, but also with charts, diagrams, and lengthy video recordings, extracting the necessary information from them.

Google says Argon is designed to support deep reasoning throughout complex workflows that may consist of a large number of sequential tasks.

Google compares Argon with the latest models from competitors

The release of Gemini 4 Argon comes amid a new stage in the race among the biggest AI developers. OpenAI recently unveiled Astra, calling it its most powerful model, while Anthropic had earlier released Fable and new versions of Opus.

Google, in turn, claims that Argon significantly outperforms GPT-6 Astra and Anthropic’s Fable and Opus models in various tests. The company also cites results from Vals, a startup focused on comparing AI models, where Argon currently holds first place in the ranking.

At the same time, such comparisons require careful interpretation: the results depend on the selected tests and evaluation methodology, and model developers themselves have an interest in demonstrating the advantages of their systems.

Google returns to the center of the AI race

Until recently, Google was considered one of the lagging participants in the race for the most powerful generative models. However, the company has recently strengthened Gemini’s position considerably.

In August, Google said that more than one billion people use the Gemini app each month. For comparison, OpenAI also said that ChatGPT had reached the milestone of one billion monthly users.

Google is now taking the next step by showing Gemini’s capabilities not only as a universal assistant, but also as a tool capable of independently carrying out complex engineering and defensive tasks.

The emphasis on cybersecurity is especially telling: a model that can find and fix vulnerabilities on its own could potentially significantly reduce the time between discovering a problem and resolving it. But at the same time, precisely these capabilities require especially careful oversight — particularly if the system gains the ability to modify software code on its own.