Google has unveiled Gemini 4 Argon, the company’s latest frontier artificial intelligence model, with a focus on cybersecurity, advanced reasoning and complex, long-duration tasks. The new model will be available first to a small group of trusted cybersecurity defenders as the company scales it down.
Gemini 4 Argon is designed to handle demanding workflows in software engineering, enterprise knowledge work, finance, legal tasks and cybersecurity.
Google says the model can sustain deeper reasoning over lengthy tasks, and it can work through complicated problems that require multiple steps instead of providing only short responses.
One of the key areas of the new model is cybersecurity, according to Google. Argon has been specifically trained to assist cyber defenders in identifying, validating and patching critical software vulnerabilities, Google says.
The company says the model has shown good performance in vulnerability remediation testing and is already being tested by cybersecurity providers in its Fairwind Program and is now being used in the cyber domain.
Wiz is one of the early users of Argon and has launched Scan for Good to identify and fix high-risk vulnerabilities affecting critical infrastructure. Argon was able to identify a serious vulnerability in software that hospitals have been using for their systems that could expose sensitive personal information, Google said.
The model also has a big increase in output capacity. Gemini 4 Argon can produce up to 1 million tokens in one output, compared to the 64,000-token limit of earlier Gemini models.
Google says the increased capacity is intended to support complex, long-running tasks that require extensive reasoning and large amounts of information.
Argon, in addition to cybersecurity, is also a powerful tool for coding and enterprise applications. Google has talked about its performance in software engineering, business automation, financial research, legal work and long-video understanding.
The company has also said that its engineers have been using the model internally for software development and other technical work.
But Gemini 4 Argon is not available yet. Google is first working with trusted cybersecurity experts and is involved in a voluntary pre-release evaluation process with the US government.
Developers, businesses and consumers will be more widely available in the future with paid API customers and Google AI Ultra subscribers being among the first groups after the cybersecurity testing phase.
Google is also emphasizing safeguards before expanding access. Argon has protections against misuse and indirect prompt injection attacks, and systems are being developed to monitor alignment problems and stop execution when necessary.
With Gemini 4 Argon, Google is putting advanced reasoning and cybersecurity capabilities at the center of its latest AI push but is in a controlled fashion to make the model more widely available.
