Google unveils Gemini 4 Argon, its most advanced AI model to date

Alphabet announced the launch of Gemini 4 Argon, its latest artificial-intelligence system, highlighting significant gains in coding, cybersecurity and complex professional workloads. The company said the model set a new record in real-world software engineering benchmarks, tied for top performance in cybersecurity tests, and led a separate evaluation covering finance, legal and other professional tasks.

Internally, Argon is already being used to optimise memory usage in Google’s data centres, freeing hundreds of terabytes without additional hardware, and has been employed by quantum-computing researchers. Google plans a phased rollout, beginning with trusted cybersecurity partners while coordinating pre-release safety reviews with the U.S. government.

"Starting this rollout in this way gives us more confidence, but also enables us to put a model that is trained and strong in cyber defense in the hands of defenders as soon as possible," said Tulsee Doshi, Google’s Gemini model product lead.

The release follows a voluntary AI-safety agreement signed by CEO Sundar Pichai with the White House. Doshi described Argon as a "well-rounded" model capable of handling long, multi-step tasks and noted it surpasses the earlier 3.8 Flash Cyber model in vulnerability discovery. She added that the model’s performance is "meaningfully important for defenders."

Google said it will extend safeguards in four key cybersecurity areas, including misuse and prompt-injection protection, before making the model publicly available.