Google has introduced Gemini 4 Argon, the most capable artificial intelligence model it has built to date, and is releasing it cautiously, starting with a small group of trusted cybersecurity partners rather than the general public.
What Google says Argon can do
Alphabet, Google’s parent company, unveiled the model on Wednesday. It described it as a significant step up in software coding, cybersecurity and complex professional work. According to the company, Argon set a new record on a benchmark of real-world software engineering, tied for first place in cybersecurity testing and led a test covering finance, law and other professional tasks.
The model can scan large codebases, isolate serious vulnerabilities and write security patches that take the surrounding code into account. It builds on Gemini 3.8 Flash Cyber, a specialised model Google released earlier this month, and outperforms it at finding weaknesses.
Google says the system is already doing useful work inside the company. Its engineers have used it for debugging and migrating codebases, and it helped optimise memory use across Google’s data centres, freeing up hundreds of terabytes without the purchase of additional hardware. Tulsee Doshi, who leads product for the Gemini models, described it as an unusually well-rounded system.
A phased release
Rather than a broad launch, Google is rolling Argon out in stages. Early access goes to cybersecurity partners, and the wider release will wait until the company has strengthened safeguards in four key areas of cyber risk. The approach reflects a concern shared across the industry: a model that is good at finding software flaws could, in the wrong hands, be used to exploit them.
The timing is notable. The launch came a day after Alphabet chief executive Sundar Pichai signed a voluntary AI safety agreement at the White House, a sign of how closely the industry and Washington are now watching frontier models.
Competition and pricing
Argon arrives almost a year after Gemini 3, which put Google back near the front of the AI race. In the months since, analysts say, the company had slipped away from the frontier, and some investors had begun to doubt whether it could build a model to match its rivals.
Morningstar said the new model leads or ties systems from OpenAI and Anthropic across a range of industry benchmarks, with the clearest gains in agentic coding, where an AI system works through long, multi-step software tasks largely on its own. Google also points to the benchmarking firm Vals, whose model index currently ranks Argon first.
Pricing is another talking point. Morningstar noted that Google’s introductory prices are significantly lower than those of rival frontier models. That could help it win business customers, and the firm argues that serving Google’s own models to clients is a more profitable use of computing power than renting capacity to outside developers.
Investors remain cautious
Markets have been less decisive. Alphabet shares fell about 2% to roughly $338.85 on Wednesday, even as the wider technology sector rose, then gained around 2% in early trading on Thursday. Some analysts are firmly positive: BNP Paribas reiterated an outperform rating with a $420 price target, citing the company’s lead in AI development and its potential consumer applications. Alphabet’s market value stands at about $4.2 trillion.
The sticking point is that the launch starts narrow. Until Argon reaches ordinary customers and developers, investors cannot judge how quickly it will translate into cloud revenue or new search products. On its second-quarter earnings call, Pichai said Google needed a larger base model to compete at the top level, and Gemini 4 is meant to be that model.
What to watch
Three questions will shape the next few weeks. The first is when Google opens Argon to the public, which depends on how quickly it can put the additional cyber safeguards in place. The second is how rivals respond on price and capability. The third is whether the model’s strong benchmark scores hold up in everyday use, where results can differ from controlled tests.
For security teams, the immediate prize is a tool that can find and fix flaws faster than human reviewers alone. For the wider industry, Argon is another reminder that the most powerful AI systems are increasingly being released in stages, with safety measures set by the developers themselves.
1 comment
[…] Google Unveils Gemini 4 Argon, Its Most Advanced… […]