Google Rolls Out Gemini 4 Argon to Cybersecurity Partners

Google is beginning a phased rollout of Gemini 4 Argon to trusted cybersecurity partners as it evaluates the model’s safety with the U.S. government and prepares for a broader launch.

Argon set a new record in real-world software engineering and tied for first in cybersecurity evaluations. It also led the Vals Index, which measures performance on professional tasks across fields including finance and law.

The model tied OpenAI’s GPT-6 Astra and Grok 4.7 in cybersecurity testing. It ranked ahead of GPT-6 Astra and Anthropic’s Fable 5.1 on the Vals Index.

Google is already using Argon to improve memory efficiency across its data centers. The work has freed hundreds of terabytes of memory without additional hardware, while quantum-computing researchers have also used the model.

Argon outperforms the 3.8 Flash Cyber model in vulnerability discovery, according to the company’s evaluation. Before a public release, Google plans to strengthen safeguards in four key areas, including misuse and prompt injection.

The launch follows Google DeepMind’s post-training work on Gemini 4, which preceded the company’s target of releasing the model well before the end of 2026.

Gemini 4 marks Google’s return to its highest-end model push after a year focused on faster, lower-cost flash models. The release comes a day after Alphabet CEO Sundar Pichai signed a voluntary AI safety accord with President Donald Trump, following a meeting with major technology executives.

Google will continue the phased rollout and safety work before making Gemini 4 Argon available more broadly.