Google's gemini 4 argon targets cybersecurity with advanced AI capabilities

Google has introduced Gemini 4 Argon, a cutting-edge AI model designed for cybersecurity and enterprise tasks, available exclusively to a select group of trusted partners through its Fairwind Program. The model, launched on Thursday, is positioned as a powerful tool for software engineering, legal, financial, and cybersecurity work. Google claims it is already transforming internal operations, with one example showing Argon agents optimizing memory usage in data centers, freeing over 300 TiB of memory. The company states that Gemini 4 Argon outperforms OpenAI's GPT-6 Astra, Claude Fable 5.1, and Claude Opus 5.5 on most benchmarks. It is trained to autonomously identify and patch critical software vulnerabilities, which is why it is offered without guardrails only to trusted defenders and Google's internal teams. Pricing starts at $2 per million input tokens and $10 per million output tokens. The release follows similar strategies by OpenAI and Anthropic, who have also limited access to their most advanced models due to cybersecurity concerns. Google highlights the model's ability to detect and address security vulnerabilities both proactively and reactively, reinforcing its role in advanced threat defense. The company's blog post details how Argon is being integrated into its internal workflows, marking a significant shift in how Google develops its products.