Google has introduced Gemini 4 Argon, its latest frontier artificial intelligence model designed for complex, long-running tasks across software engineering, enterprise knowledge work and cybersecurity.
The model is initially being rolled out to a select group of cybersecurity partners through Google's Fairwind Program rather than receiving an immediate broad release. Google said it is taking a phased approach while collecting feedback and testing safeguards before expanding Argon to developers, enterprises and consumers.
One of Argon's key areas of focus is defensive cybersecurity. According to Google, the model has been trained to autonomously identify, validate and patch critical software vulnerabilities. The company is positioning the capability for trusted cyber defenders as AI models become increasingly capable of handling multi-step security tasks.
Beyond cybersecurity, Argon has been developed for coding, research, writing and other professional workflows. Google said thousands of its employees are already using the model internally for specialised coding tasks, research and engineering work.
The company highlighted several internal examples, including codebase migrations and memory optimisation across its data centres. In one project, Argon agents were used to migrate C and C++ codebases to Rust, ranging from smaller libraries to more than 800,000 lines of code for the Zircon kernel used by Google's Fuchsia operating system. Google said such migrations continue to undergo automated and manual review before being deployed.
Argon also comes with a one million-token context limit, allowing the model to work with large amounts of information during multi-step tasks. Its capabilities extend to enterprise knowledge work, including legal and financial research, as well as visual analysis involving videos and charts.
Google is also pitching Argon as a step forward in AI reasoning. The company said the model performed strongly across several external benchmarks and cited testing that placed it ahead of competing frontier models in some evaluations. As with other benchmark claims made by AI companies, results can vary depending on the tests and methodology used.
Pricing is expected to begin at an introductory rate of $2 per million input tokens and $10 per million output tokens, before increasing to $4 and $20 respectively after the introductory period.
Google said wider availability will begin with paid API customers and Google AI Ultra subscribers after its initial testing phase. The release adds another frontier model to a market where Google, OpenAI, Anthropic and other AI developers are competing across reasoning, coding, agents and enterprise applications.
Disclaimer: This article may include information derived from interviews, press releases, public statements, research, company communications and other publicly available or third-party sources. Such material may be summarised, paraphrased or contextualised for journalistic and editorial purposes. All rights in third-party content remain with their respective owners.