alphabet on Wednesday unveiled Gemini 4 Argon, its most advanced artificial intelligence model to date, delivering significant improvements in coding, cybersecurity, and complex professional tasks.
The company says the model has set new records in real-world software engineering, ranks first in cybersecurity, and leads other benchmarks measuring performance across finance, legal, and other professional functions.
Argon is already being used internally to optimize memory in Google’s data centers, and the company said it can free up hundreds of terabytes of memory without purchasing additional hardware. Quantum computing researchers also use this model.
Google plans to launch the new model in phases, starting with trusted cybersecurity partners, while working with the U.S. government to conduct pre-release safety assessments.
Tulsee Doshi, Google’s Gemini model product lead, told CNBC: “Starting the rollout this way not only gives us more confidence, but also allows us to get trained, cyber-hardened models into the hands of defenders as quickly as possible.”
The release of the model comes a day after CEO Sundar Pichai signed a voluntary agreement at the White House with President Donald Trump and executives from major technology companies after a luncheon to address growing concerns about the safety of AI.
Gemini 4 also launches nearly a year after Gemini 3 brought tech companies back to the forefront of the heated AI model race. This long-awaited model represents a major step forward into the frontier and aligns with Google’s commitment to expanding its faster, lower-cost flash model.
Google said Wednesday that it is considering expanding safeguards in four key cybersecurity areas, including exploitation and instant injection, before publicly launching.
Doshi described the Gemini 4 as an “incredibly balanced” model that excels at performing long, multi-step tasks. This new model is also a significant step up in cybersecurity from Google’s 3.8 Flash Cyber model, released earlier this month, and offers superior performance in finding vulnerabilities.
“We truly believe that a model of this caliber and this level of frontier performance has meaningful importance to the Defender,” Doshi said.
Argon is on par with OpenAI’s GPT-6 Astra and Grok 4.7 on cybersecurity benchmarks, and outperforms GPT-6 Astra and recent Anthropic models on the Vals index.
WATCH: Google’s head of Gemini product details weeks of company-wide testing ahead of Gemini 4 Argon launch
