Google's Gemini 4 Argon: A Powerful New AI Model for Cybersecurity and Beyond

Google's latest AI model, Gemini 4 Argon, has been launched to a select group of trusted cyber defenders through Google's Fairwind Program. The frontier AI is considered one of the most capable models yet, outperforming OpenAI GPT-6 Astra and Anthropic's Claude Fable 5.1 on most benchmarks. According to Google, Gemini 4 Argon excels at software engineering, enterprise-class legal and financial work, and cybersecurity tasks.

One example of its capabilities is demonstrated in a blog post by Google, where a team of Argon agents analyzed memory usage in Google's data centers and optimized it, freeing over 300 TiB (tebibytes; one TiB is roughly 1.1 terabytes) of memory. The model has also been trained to be particularly capable in cybersecurity defense, able to autonomously find, validate, and patch critical software vulnerabilities.

However, due to its powerful capabilities, Google is releasing Gemini 4 Argon without guardrails only to trusted defenders and internal teams. This decision follows OpenAI and Anthropic's similar moves with their most capable models. The pricing for the model will be introductory prices of $2 per million input tokens and $10 per million output tokens.