CALIFORNIA / RankWire.AI / – Google has unveiled Gemini 4 Argon, its latest premier artificial intelligence model aimed at complex professional applications. The company announced Argon on Sept. 30 as the leading model within the Gemini 4 series. It is capable of supporting intricate software development, financial analysis, legal research, and cybersecurity tasks. Additionally, the model manages extended sequences of reasoning and execution. Currently, access is restricted to select users, with cybersecurity defenders participating through the Fairwind Program.
The maximum output capacity of Argon has been increased to 1 million tokens, representing a significant jump from the previous 64,000-token limit. This enhancement enables the model to handle longer projects without the need to split work across multiple sessions. The introductory API pricing begins at $2 per million input tokens, while output tokens are priced at $10 per million during this period. Inputs stored in cache benefit from a 95% discount. Future pricing will shift to $4 for input tokens and $20 for output tokens.
Numerous employees within Google are already utilizing Argon for tasks such as coding, research, and writing. Internal testing has involved data center optimization and large-scale software migration efforts. One example included using Argon agents to facilitate C and C++ migrations to Rust. Another project focused on memory profiling across data centers, which resulted in freeing over 300 tebibytes of memory. Further savings were identified through ongoing analysis of these systems.
Enhanced capabilities for extensive technical tasks
Google reported a 77.9% performance score for Argon on DeepSWE v1.1, a benchmark for evaluating long-term software engineering skills. Results for finance, legal tasks, automation, and multimodal functions have also been released. Argon was developed by Google DeepMind as part of the broader Gemini model family, which integrates coding tools with long-context reasoning and multimodal functionalities. Its expanded output capacity is designed to support workflows that involve multiple interconnected steps before completing a task.
Cybersecurity remains at the forefront of the initial deployment. Argon can detect, verify, and address software vulnerabilities within authorized defensive environments. Wiz is leveraging the model through its Scan for Good initiative, which aims to identify security flaws in public infrastructure. Google reported a 68% score on CWE-bench v1, a benchmark dedicated to vulnerability remediation. Selected security defenders can also utilize Argon without standard cyber guardrails when working on approved defensive security operations.
Limited public access during staged rollout
A definitive release date for broad public availability of Gemini 4 Argon has not been announced by Google. The company is employing a phased approach, collecting feedback from early users, and participating in a voluntary U.S. government process that grants pre-release access to advanced AI models. Later, access will expand to developers, enterprise clients, and consumers. Priority for initial access will go to paid API users and Google AI Ultra subscribers, though an exact launch date remains undisclosed.
It was also confirmed that Gemini 3.5 Pro will not be released. This model was expected prior to the Gemini 4 series. Instead, Argon now represents the flagship offering tailored for demanding reasoning tasks and professional workflows. Other Gemini models are still available to meet different performance and pricing requirements. For now, Gemini 4 Argon remains in the hands of trusted testers, cybersecurity partners, and early-access programs, with wider release still pending.