Google has officially introduced Gemini 4 Argon, a new advanced AI model designed to handle complex reasoning, long-running tasks, software development, and professional workflows, with a strong focus on cybersecurity.
The new model is part of Google’s latest generation of Gemini systems and is built to go beyond simple question answering by completing more complicated, multi-step tasks over extended periods.
Designed for Long and Complex Tasks
One of the biggest improvements in Gemini 4 Argon is its ability to work through large and complicated tasks without losing context.
Google says the model supports up to 1 million output tokens, giving it significantly more room to generate, analyze, and process information during large-scale workflows.
This could be especially useful for developers, researchers, and businesses working with large documents, databases, codebases, and projects that require extensive analysis.
A Strong Focus on Software Development
Software engineering is one of the main areas where Google is positioning Gemini 4 Argon.
According to the company, Google employees are already using the model for debugging, algorithm design, software optimization, and large-scale programming tasks.
Google also says Argon has been used to help migrate software projects from languages such as C and C++ to Rust.
In internal and external benchmarks, the model achieved a 77.9% score on DeepSWE v1.1, a benchmark designed to evaluate AI systems on realistic, long-horizon software engineering tasks.
Google also reported using Argon-powered agents to analyze internal systems and improve memory efficiency across some of its infrastructure.
Built for Professional and Enterprise Workflows
Gemini 4 Argon is not limited to software development.
Google says the model also performs strongly across several professional domains, including:
- Financial research and analysis
- Legal research and document drafting
- Tax-related workflows
- Data and chart analysis
- Long-form video understanding
- Multi-step business processes
The company says Argon scored 51.3% on AutomationBench, a benchmark from Zapier designed to measure whether AI systems can successfully complete end-to-end business tasks.
Advanced Cybersecurity Capabilities
Cybersecurity is another major focus of Gemini 4 Argon.
Google says the model has been trained to identify software vulnerabilities, validate security issues, and assist with fixing them in a more autonomous way.
During testing, the model reportedly discovered vulnerabilities across complex software projects using around 20 different programming languages.
Gemini 4 Argon also achieved a score of approximately 68% on CWE-bench v1, a benchmark used to evaluate AI systems on vulnerability detection and remediation tasks.
Google is initially giving selected cybersecurity experts access to the model through its Fairwind program before making it more widely available.
Google Is Taking a Gradual Approach to the Launch
Despite its advanced capabilities, Gemini 4 Argon is not being released broadly to everyone immediately.
Google says it is using a gradual rollout strategy because of the model’s powerful capabilities, especially in areas such as cybersecurity.
The company is also working on additional protections designed to reduce the risk of misuse and improve the model’s resistance to attacks such as prompt injection, where attackers attempt to manipulate an AI system into ignoring its original instructions.
How Much Will Gemini 4 Argon Cost?
Google has also revealed preliminary pricing for Gemini 4 Argon through its APIs.
During the introductory period, pricing will be:
$2 per million input tokens and $10 per million output tokens.
Cached input tokens will cost significantly less, with Google offering a discount of up to 95% compared with standard input pricing.
After the introductory period ends, Google plans to increase pricing to:
$4 per million input tokens and $20 per million output tokens.
When Will Gemini 4 Argon Be Available?
For now, Gemini 4 Argon is being released to a limited group of trusted testers and cybersecurity researchers.
Google plans to expand availability gradually, starting with paid API customers and Google AI Ultra subscribers before potentially making the model available to a broader audience.
Gemini 4 Argon represents another step in Google’s effort to build AI systems that can do more than simply respond to prompts.
Instead, the company is increasingly focusing on models capable of completing long, complex projects across software development, research, cybersecurity, and professional business workflows.
As competition in the AI industry shifts from conversational assistants toward more autonomous AI agents, Gemini 4 Argon could become one of Google’s most important models in this new generation of artificial intelligence.
