Alphabet unveiled Gemini 4 Argon on Wednesday, describing it as its most advanced artificial intelligence model yet, with major improvements in software engineering, cybersecurity and complex professional work. Google plans to introduce the model in stages, initially making it available to trusted cybersecurity partners while conducting pre-release safety evaluations with the US government.
The company said Gemini 4 Argon set a new record in real-world software engineering and tied for first place on cybersecurity benchmarks. It also led another benchmark assessing performance across professional fields including finance and law. The results place the model among the leading systems in several areas of advanced AI performance.
Argon is already being used internally at Google, where the company said it has helped optimise memory at its data centres. The work has freed hundreds of terabytes of memory without requiring the purchase of additional hardware. Quantum computing researchers have also used the model, according to Google.
Tulsee Doshi, Google’s Gemini model product lead, said the staged release was intended to balance access to the model’s capabilities with safety precautions, particularly because of its cybersecurity performance. “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 defence in the hands of defenders as soon as possible,” Doshi told CNBC.
Google said it was working to scale safeguards in four key cybersecurity areas, including misuse and prompt injection, before making the model publicly available. Doshi described Gemini 4 as an “incredibly well-rounded” model that is particularly capable of handling long, multi-step tasks.
The new system also represents a significant cybersecurity improvement over Google’s Gemini 3.8 Flash Cyber model, which was released earlier this month, with Argon outperforming that model in vulnerability discovery. “We really believe that a model of this caliber and this level of frontier performance is meaningfully important for defenders,” Doshi said.
On cybersecurity benchmarks, Argon ties with OpenAI’s GPT-6 Astra and Grok 4.7, while ranking ahead of GPT-6 Astra and recent Anthropic models on the Vals Index, according to the source.
The launch came a day after Google CEO Sundar Pichai signed a voluntary AI safety agreement with US President Donald Trump and other major technology executives at the White House. The agreement followed a luncheon focused on rising concerns over AI safety.
Gemini 4 arrives nearly a year after Gemini 3, as Google continues its push to compete at the frontier of AI models. The company’s latest release follows a strategic shift towards scaling faster, lower-cost Flash models while continuing to develop more advanced systems.
The initial availability of Gemini 4 Argon will therefore be restricted to trusted cybersecurity partners, with Google conducting further safety work before a broader public launch. The company has not yet released the model generally.
