Google has introduced Gemini 4 Argon, a new artificial intelligence model designed to push the limits of long and complex workflows. The company is positioning it primarily for coding, professional data analysis, and cybersecurity. However, the model is not yet available to the general public.
Gemini 4 Argon was built to handle tasks that require multiple consecutive steps and extended reasoning. According to Google, the model can be used for software development, financial analysis, legal research, and working with large documents or long-form video content.
Google is already using Argon internally. The model has helped teams migrate large portions of code from C and C++ to Rust. In one case involving the libgav1 video decoder, Argon reportedly rewrote tens of thousands of lines of code, producing a version that was 2.7 times faster than an earlier Rust port.
Other teams have used AI agents powered by Argon to identify potential memory savings in Google’s data centers.
Google Is Focusing Heavily on Cybersecurity
Cybersecurity played a major role in the development of Gemini 4 Argon. The model is designed to autonomously identify, validate, and patch critical software vulnerabilities.
For now, Google is giving selected security researchers and partners access to its most advanced cybersecurity capabilities through the Fairwind Program.
Cybersecurity company Wiz is already using Argon as part of its Scan for Good initiative, which searches for vulnerabilities in critical infrastructure free of charge. According to Google, the model identified a serious vulnerability in software used by hospitals that previous leading AI models had failed to detect.
On CWE-bench v1, a benchmark designed to test the ability of AI systems to repair software vulnerabilities, Argon achieved a score of 68%.
The model also appears to perform strongly outside cybersecurity. Independent benchmarking platform Vals currently ranks Gemini 4 Argon first on its main Vals Index with a score of 68.9%. Argon also took the top position in the Finance Agent benchmark, which evaluates models on multi-step financial research tasks.
Google Is Keeping Argon Under Tight Control for Now
The wider public cannot access Gemini 4 Argon yet. Google is opting for a gradual rollout while it continues testing safeguards against misuse, prompt injection attacks, and potentially unwanted autonomous behavior.
After the initial phase with selected cybersecurity partners, access is expected to expand to a broader group of developers and companies, followed later by Google AI Ultra subscribers.
Pricing will also be closely watched. Google announced an introductory rate of USD 2 per million input tokens and USD 10 per million output tokens. Once the introductory period ends, those prices are expected to double to USD 4 and USD 20 respectively.
Gemini 4 Argon marks another step in the rapidly intensifying competition among the world’s leading AI laboratories. Early benchmark results suggest that Google has produced a model capable of competing at the highest level across coding, professional research, and cybersecurity.
