CALIFORNIA / RankWire.AI / – Google has revealed its latest flagship AI model, Gemini 4 Argon, designed for tackling complex professional tasks. Announced on Sept. 30, this model marks the core of the new Gemini 4 series. It is optimized for roles in software development, financial analysis, legal work, and cybersecurity defense, with Google noting its enhanced ability to perform deeper reasoning across multi-step, extended tasks. Access is currently limited to a select group of cybersecurity defenders via the Fairwind Program.
The maximum output capacity of Gemini 4 Argon has increased to 1 million tokens from the previous 64,000, enabling the model to handle longer processes within a single session. Google’s initial API pricing has been set at $2 per million input tokens and $10 per million output tokens, with cached input tokens receiving a 95% discount. Following the initial phase, prices are scheduled to increase to $4 and $20, respectively.
According to Google, thousands of employees are already leveraging Argon for specialized coding, research, and writing tasks. Internal teams have also applied the model to optimize data center memory and migrate large codebases, including a project using Argon agents to migrate C and C++ code to Rust. Another effort involved deploying agents to analyze memory usage across Google data centers, resulting in over 300 tebibytes of freed memory, with additional savings identified through ongoing work.
Enhanced capabilities for demanding professional tasks
Google reported a 77.9% score for Gemini 4 Argon on DeepSWE v1.1, a benchmark for long-term software engineering performance. The model has also demonstrated strong results in financial, legal, and automation assessments. Supporting multimodal reasoning alongside coding and enterprise workflows, Argon is part of the broader Gemini family developed by Google DeepMind. The increased output capacity is designed to manage extended workflows that require multiple sequential reasoning and execution steps.
Cybersecurity remains a key focus in the initial deployment of the model, with Google stating Argon can identify, verify, and patch software vulnerabilities in controlled defensive environments. Through its Scan for Good initiative, Wiz is utilizing Argon to detect security vulnerabilities in public infrastructure. Google reported that Argon achieved a 68% score on CWE-bench v1, a benchmark for vulnerability remediation, and is providing selected cybersecurity defenders with access to the model outside its standard guardrails for approved security work.
Limited launch before broader availability
No specific date has been announced for the public release of Gemini 4 Argon, as Google states it is adopting a phased rollout and gathering feedback from early testers. The company also participates in a voluntary U.S. government process for pre-release model access. Future availability is expected to extend to developers, businesses, and consumers, beginning with paid API clients and Google AI Ultra subscribers, although no fixed launch date has been provided for these groups.
Furthermore, Google has confirmed that Gemini 3.5 Pro, originally scheduled for June, will not be released. Consequently, Gemini 4 Argon remains the company’s leading flagship model for complex reasoning and professional tasks. While other Gemini models are still available for different performance and budget needs, Argon introduces a larger output capacity, advanced coding features, and specialized cybersecurity capabilities. Currently, its access is limited to trusted testers and selected security partners involved in defense-related projects.