Google (GOOGL.US) Gemini 4 Argon makes a big debut: it claims to be “the most advanced so far”, and the AI competition heats up again after reaching the top of multiple benchmarks

The Zhitong Finance App learned that Alphabet (GOOGL.US) officially released Gemini 4 Argon on Wednesday. This is its most advanced artificial intelligence model to date, which has achieved major upgrades in programming, cybersecurity, and complex professional work. According to the company, the model broke records in real-world software engineering tests, tied for first place in cybersecurity benchmarks, and topped another benchmark that measures performance in financial, legal, and other professional tasks.

Google said that Argon is already being used on the intranet to optimize data center memory, freeing up hundreds of terabytes of memory without purchasing additional hardware. Quantum computing researchers have also used this model.

Google plans to launch the new model in stages, first targeting trusted cybersecurity partners, and cooperating with the US government to conduct pre-launch security assessments.

Tulsee Doshi, head of Google's Gemini model product, said, “Starting the promotion in this way gives us more confidence and allows us to hand over a trained model with strong cyber defense capabilities to defenders as soon as possible.”

The day before this announcement, Google CEO Sundar Pichai signed a voluntary agreement with US President Donald Trump and key tech executives at the White House. Earlier, all parties held a luncheon to discuss rising AI security concerns.

Gemini 4 was released almost a year after Gemini 3. Gemini 3 brought the tech company back to the cutting edge of the heated AI model competition. This highly anticipated new model represents a major step forward for Google in the cutting edge field, and also continues its strategy to move to a faster, lower-cost Flash model.

Google said on Wednesday that it is seeking additional security protections in four key cybersecurity areas, including abuse and tip injection, prior to public launch.

Doshi called Gemini 4 an “extremely comprehensive” model that excels at performing long, multi-step tasks. The new model is also significantly upgraded in terms of network security compared to the 3.8 Flash Cyber model released by Google earlier this month, and performs better in finding vulnerabilities.

“We do believe that a model with this level and cutting-edge performance is important to the defender,” Doshi said.

In cybersecurity benchmarks, Argon tied with OpenAI's GPT-6 Astra and Grok 4.7; ahead of recent GPT-6 Astra and Anthropic models in the Vals Index.