Liam Walsh
Liam is a Co-Director at Intelligency and heads up the agency's Digital Intelligence & Paid Social activity. Over the last decade, he has worked with brands from the world of sports such as Premier League clubs to entertainment such as Channel 4 and Disney.
Google has announced Gemini 4 Argon, its most capable AI model to date, and what sets it apart from previous releases is not just raw performance but where that performance is being directed. Rather than a general-purpose upgrade, Argon has been built specifically to handle the kinds of complex, multi-step tasks that businesses struggle with, such as large-scale software development, financial research, legal work, and cybersecurity defence. It is currently rolling out to a select group of trusted testers, with broader availability for developers, enterprise customers and consumers to follow.
Built for Real Work, Not Just Benchmarks
The headline capability is an industry-leading output limit of one million tokens, a significant leap from the previous 64,000. In practical terms, this means the model can sustain deep, uninterrupted reasoning across far longer and more complex tasks without losing context. Google has already put Argon to work internally, where it has been used to migrate large codebases, optimise memory usage across data centres, and assist quantum computing researchers, in one case beating a published performance baseline by 40% in minutes. On AutomationBench, Zapier’s benchmark measuring execution across core business functions, Argon ranks first with a score of 51.3%. It also leads on benchmarks covering financial research, legal drafting and long-form video understanding.
A New Frontier for Cybersecurity
One of Argon’s most significant capabilities is its ability to autonomously identify, validate and patch software vulnerabilities, something previous models have not been able to do reliably. Google has already deployed it through its Fairwind Program with cybersecurity firm Wiz, where it uncovered a critical vulnerability in healthcare software used by hospitals worldwide, a risk that earlier frontier models had missed entirely. For organisations managing complex digital infrastructure, this represents a meaningful shift in what AI-assisted security can deliver in practice.
What This Means for Businesses Watching the AI Space
Argon is not yet publicly available, and Google is being deliberately cautious about the rollout, investing in safeguards against misuse before opening broader access. Pricing starts at $2 per million input tokens and $10 per million output tokens at the introductory rate. For businesses tracking how AI is reshaping the tools they rely on, from campaign automation to data analysis to content workflows, Argon signals the direction of travel clearly. The gap between what AI can handle and what previously required specialist human expertise is closing faster than most organisations have planned for.
