Google has unveiled a new AI model. Will it help the company regain its lead in the AI race?

Google has unveiled Gemini 4 Argon—its most advanced artificial intelligence model. The company claims it significantly outperforms its predecessors in programming, cybersecurity, and performing complex professional tasks, including those in the legal and financial sectors. It calls Argon the beginning of a “new era of cutting-edge intelligence.”

Gemini 4 Argon delivered industry-leading results on several benchmarks, outperforming its competitors. In one test evaluating the ability to detect and fix vulnerabilities, it tied for first place with OpenAI’s GPT-6 Astra, according to a press release. “We truly believe that a model of this caliber, with such advanced capabilities, is of significant importance to cybersecurity professionals,” Tulsi Doshi, Gemini product lead at Google, told CNBC.

The new project will be rolled out in phases: initially, it will be made available to Google’s trusted cybersecurity partners, and at the same time, the company will work with the U.S. government to assess the risks. The company did not specify when Argon will become publicly available, noting only that it intends to strengthen security measures—including protections against abuse and attacks—before the public release. In addition, Google has made an effort not to use the monitoring results in further training of the model, so as not to teach Argon how to bypass control systems, the tech giant said.

Following the announcement of the promotion, shares of Google's parent company, Alphabet, rose 1.5% in after-hours trading.

The success of the new Gemini series is extremely important for Google, Bloomberg notes. Last November, the company unveiled Gemini 3. The market welcomed this release, viewing it as a turning point in its efforts to catch up with OpenAI and Anthropic. In May, Google announced a new version, Gemini 3.5 Pro, and promised to release it the following month. Since then, the release date has been pushed back repeatedly, and ultimately, the company abandoned the project altogether, sources told the agency. According to Bloomberg Intelligence analyst Mandip Singh, training such a model could have cost up to $400 million.

The long-awaited new advanced model could help Google regain its leading position in the race to develop AI for mission-critical tasks, including security and programming, according to Business Insider. Although the company has released several updates to its Flash model lineup, it has fallen behind its competitors—OpenAI and Anthropic—in areas such as software development and cybersecurity, the publication writes.

However, Bloomberg reports skepticism regarding Argon’s prospects. According to the agency’s sources, who are familiar with the results of internal testing, its programming capabilities are inconsistent: despite strong benchmark results, it performs worse on certain tasks in real-world use. In addition, a Bloomberg source noted that it is a very large model that is expensive to operate, which could potentially put pressure on Google’s profit margins.

That said, Gemini 4 Argon does have its strengths, according to the agency’s source. He said the model is particularly good at handling non-text data—for example, extracting metadata from videos. He also highlighted the model’s security and its capabilities in the field of cybersecurity.

Various versions of Gemini underpin nearly all of the tech giant’s products—from AI-powered responses in Google Search results, the company’s main source of revenue, to Maps, Gmail, and Chrome, Bloomberg notes. Each of these products has over a billion users, giving Google a distribution advantage that some of its competitors lack. However, OpenAI and Anthropic are increasingly shifting from selling models to creating their own products, including AI agents for programming, the agency notes. Whether Gemini 4 proves to be competitive could determine whether its rivals will have extra time to attract consumers, developers, and businesses to their platforms, Bloomberg writes.

As Google stated, it would be incorrect to claim that Gemini 4 does not perform well enough in areas such as programming, the agency adds.

Researchers whom Bloomberg had previously interviewed attributed the problems to excessive bureaucracy within the company, which is attempting to integrate AI into virtually all of its products. According to them, constantly changing requirements and priorities are also preventing Google from focusing on a unified strategy.

At the same time, a whole group of leading AI researchers left Google, including renowned engineer Jeff Dean, Nobel Prize laureate John Jumper, and Noam Shazir, who helped develop the technology underpinning much of the AI boom. In August, Demis Hassabis, who had long led the company’s AI research, assumed a new role as chairman and handed over the day-to-day management of DeepMind to his longtime deputy, Kora Kavukcuoglu.

This article was AI-translated and verified by a human editor