- Published on
Google Launches Gemini 4 Argon Model, Challenges GPT-6 Astra and Claude Opus 5.5 as Prices Drop Sharply
- Authors
- Name
- DeepHorus
Google has released its flagship artificial intelligence model, Gemini 4 Argon, joining the competition with OpenAI and Anthropic's next-generation frontier models. Google says Argon is designed for complex, long-term tasks, covering software development, enterprise knowledge work, and cybersecurity defense, with a particular focus on improving long-horizon reasoning capabilities. Its goal is to compete in the high-end model market with stronger performance and lower prices.
Argon is not yet broadly available. It is being offered first to selected cybersecurity defense organizations participating in the GoogleFairwind program for testing. Google is still conducting additional security evaluations and proceeding with deployment on a limited basis. A broader availability timeline has not yet been announced.
Compared with the previous Gemini series, Argon's maximum output length for a single request has increased from 64,000 tokens to 1 million tokens. The model can perform longer reasoning within a single task and can handle complex software engineering, research, and large-scale workflows without requiring frequent interruptions. This helps reduce the need to repeatedly restart complex tasks.
Tests published by Google show that Argon surpasses GPT-6 Astra and Claude Opus 5.5 in multiple areas. It achieved 77.9% on the DeepSWE v1.1 long-horizon software engineering test, scored 51.3% on the Zapier AutomationBench automation-task benchmark, and reached 91.7% on the LVBench long-video understanding test.
In the Vals Index test, which evaluates the ability to perform real-world economic activities such as finance, law, taxation, and software development, Google also says that Argon ranks among the leading models. Thousands of Google employees have already used the model for programming, research, and document writing. Internal use cases include quantum algorithm optimization, data-center memory optimization, and migrating large C and C++ codebases to Rust.
Google says Argon can autonomously identify, validate, and fix software vulnerabilities. The Wiz security team has already used it for security research through the "Scan for Good" program. In the CWE-bench v1 vulnerability-remediation test, Argon achieved 68%, tying with the best performance in the industry.
Argon's initial pricing is $2 per 1 million input tokens and $10 per 1 million output tokens, with cached input receiving a 95% discount. This pricing is significantly lower than that of some high-end competing models, enabling Google to strengthen its market competitiveness. However, the model remains available only on a limited basis until additional security testing and validation are completed, and its broader release timetable has not yet been announced.
