CALIFORNIA / RankWire.AI / – Google has introduced Gemini 4 Argon, its latest flagship artificial intelligence model designed for tackling complex professional workloads. Announced on Sept. 30, this model forms the core of the Gemini 4 series. Argon aims to serve fields such as software engineering, finance, legal expertise, and cybersecurity defense. Google highlighted that the model can maintain more profound reasoning capabilities during extended, multi-step tasks. The company has begun offering limited access to cybersecurity experts through its Fairwind Program.
Gemini 4 Argon significantly increases the maximum output capacity to 1 million tokens, up from the previous limit of 64,000 tokens. This enhancement enables the model to process lengthy tasks within a single operational thread. Initial API pricing is set at $2 per million input tokens and $10 per million output tokens. Input tokens stored in cache benefit from a 95% discount on the input price. After the initial phase, pricing will rise to $4 and $20, respectively.
According to Google, thousands of its employees are already leveraging Argon across specialized coding, research and writing tasks. Internal teams have also utilized the model for data center memory optimization and large-scale code migration projects. One such effort involved deploying Argon agents for C and C++ to Rust migrations, while another focused on memory profiling across Google’s data centers. Google reported that these optimization efforts freed more than 300 tebibytes of memory, with additional savings identified through ongoing work.
Enhanced capacity for intricate professional functions
Google shared that Gemini 4 Argon achieved a 77.9% score on DeepSWE v1.1, a benchmark assessing long-term software engineering performance. The firm also emphasized positive results in finance, legal tasks, and automation benchmarks. The model supports multimodal reasoning in addition to coding and enterprise workflows. Developed by Google DeepMind as part of the larger Gemini family, Argon’s increased output capacity is designed to handle extended processes that involve multiple reasoning and execution stages.
Cybersecurity remains a key focus in the model’s initial deployment. Google stated that Argon can identify, verify, and patch software vulnerabilities within controlled defense scenarios. Wiz is employing Argon through its Scan for Good initiative, which aims to detect security vulnerabilities in public infrastructure. The company reported that Argon achieved a score of 68% on CWE-bench v1, a benchmark for vulnerability remediation. Selected cybersecurity defenders are being granted access to the model without the usual guardrails for approved security tasks.
Selective introduction before broader Gemini 4 rollout
Google has not yet announced a specific launch date for the general public release of Gemini 4 Argon. The company explained that it is employing a phased rollout approach, gathering feedback from early testers. It also participates in a voluntary pre-release process with the U.S. government for model access. Future plans include extending availability to developers, enterprises, and consumers. The initial rollout will begin with paid API users and Google AI Ultra subscribers, though no fixed launch date has been provided for these groups.
Additionally, Google clarified that it has no plans to release Gemini 3.5 Pro, which was originally scheduled for June. This decision leaves Gemini 4 Argon as the company’s current flagship model for demanding reasoning and professional tasks. The company continues to offer other Gemini variants tailored to different performance levels and cost considerations. Argon distinguishes itself with its larger output capacity, advanced coding capabilities, and dedicated cybersecurity functions. Its current access is limited to trusted testers and selected security partners involved in defensive operations.