Google bets on Gemini 4 Argon for cybersecurity

Alphabet has introduced Gemini 4 Argon, an AI model trained for cybersecurity and already used internally. Google claims it outperforms benchmarks against models from OpenAI and Anthropic, while deployment begins with a limited circle of partners.

Google Claims Lead on Benchmarks Against OpenAI and Anthropic

Google asserts that Argon has achieved significantly higher scores than OpenAI's GPT-6 Astra and Anthropic's Fable and Opus across various artificial intelligence benchmarks. The company cites the startup Vals, which specializes in evaluating AI models, to indicate that Argon currently holds the top position on its index. Meanwhile, OpenAI has launched Astra, touted as its best model to date, while Anthropic released Fable earlier this year with similar claims about its performance.

Cybersecurity as a Priority and Announced Multimodal Capabilities

According to Google, Argon is designed to excel in cybersecurity and has been specifically trained for defensive tasks. The company claims that the model can autonomously detect, validate, and rectify critical software vulnerabilities. Google also highlights Argon's ability to analyze visual content, whether it be lengthy videos or graphics.

Developer Use Cases and Promise of Long-Term Reasoning

Developed by Alphabet, Argon is intended for programming, research, and writing. Google indicates that the model performs well in programming and engineering, noting that its employees are already using it for tasks such as debugging and code migrations. In a blog post published Wednesday, the company explains that Argon is designed to support deep reasoning over complex and extended workflows, which it believes is changing the way people work at Google.

Restricted Deployment and Consumer Assistant Market

Argon is currently being deployed to a limited group of specialized partners through the Fairwind Program, a Google security initiative. Additionally, the company announced in August that its Gemini application surpassed one billion monthly users, a milestone also recently reached by OpenAI's ChatGPT. Leading AI labs are ramping up the launch of increasingly powerful models while warning of the risk that these technologies could become uncontrollable.