Google has announced Gemini 4 Argon, a new frontier AI model designed for complex, long running tasks. The company says Argon can handle advanced software engineering, enterprise research, legal work, finance tasks, and cybersecurity defense.
Google is starting with limited access through its Fairwind Program. The company says it wants feedback from trusted users before expanding access to developers, businesses, and consumers.
Quick Fact
| Detail | Information |
|---|---|
| What happened | Google announced Gemini 4 Argon |
| Launch status | Limited rollout through the Fairwind Program |
| Focus | Coding, enterprise AI, visual analysis, and cybersecurity |
| Input price | $2 per million tokens |
| Output price | $10 per million tokens |
| DeepSWE v1.1 | 77.9% |
| LVBench | 91.7% |
| AutomationBench | 51.3% |
| CWE Bench v1 | 68% |
| Cybersecurity partner | Wiz, through its Scan for Good initiative |
Gemini 4 Argon Targets Complex AI Work
Gemini 4 Argon specializes at activities that involve several stages and continuous thinking. Google claims its developers are already using the model for everyday debugging, massive codebase migrations, and algorithm creation.
On DeepSWE v1.1, the company reports a score of 77.9%. This benchmark tests performance on real-world software engineering activities that may involve numerous phases.
Argon also targets corporate users. Google claims that the model tops the Vals Index, which assesses economic effect in fields such as banking, coding, legal work, and tax activities.
Leading Role in Cybersecurity Defense
Gemini 4 Argon also places a strong emphasis on cybersecurity.
Google claims that Argon can detect, test, and resolve significant software problems. The business intends to provide trusted cyber defenders and internal Google teams access to the model without the typical cybersecurity constraints.
This technique provides security teams access to the model’s whole defensive capabilities. Google says the idea is to assist defenders in identifying and addressing hazards before attackers can exploit them.
Wiz has already begun testing Argon through its Scan for Good project. The initiative focuses on identifying and addressing major security vulnerabilities impacting key public infrastructure.
Google has released numerous important findings from its internal and external reviews.
• Argon achieved 77.9% on DeepSWE v1.1.
• Argon scored 51.3% on AutomationBench.
• Argon scored 91.7% on LVBench.
• According to Google, Argon tied for first place with a score of 68% on CWE Bench v1.
• Internal testing identified vulnerabilities spanning complicated codebases and 20 programming languages.
• Wiz’s internal testing revealed that Argon outperformed Gemini 3.8 Flash Cyber in attack surface discovery, vulnerability identification, and proof-of-concept validation.
These data are based on the announcement and evaluations given by Google.
Gemini 4 Argon Pricing and Rollout
Google has also announced introductory pricing of $2 per million input tokens and $10 per million output tokens. Cached input tokens will cost 95% less than the standard input price.
Gemini 4 Argon marks Google’s latest push toward AI systems that can handle complex, long running workflows.
Its reported results cover software engineering, enterprise work, visual understanding, and cybersecurity. The initial rollout remains limited, with Google planning to expand access after further testing and feedback.
