CALIFORNIA / RankWire.AI / – Google has unveiled Gemini 4 Argon, its latest flagship AI model designed for sophisticated professional applications. Announced on Sept. 30, Argon is positioned as the leading model in the Gemini 4 series. It is capable of supporting complex tasks such as software development, financial analysis, legal research, and cybersecurity operations. The model also excels in managing extended sequences of reasoning and execution. Currently, access remains restricted, with selected cybersecurity experts utilizing Argon through the Fairwind Program.
The maximum output limit for Argon has been increased to 1 million tokens, a significant jump from the previous cap of 64,000 tokens. This enhancement enables the model to undertake longer and more demanding tasks without fragmenting work into multiple sessions. Initial API pricing begins at $2 per million input tokens, with output tokens costing $10 per million during the same period. Inputs stored in cache benefit from a 95% discount. Future pricing adjustments are expected to set the rates at $4 for input tokens and $20 for output tokens.
Numerous employees within Google are already leveraging Argon for activities like coding, research, and content creation. Internal teams have put the model to test in data center optimization and large-scale software migration projects. One such project involved deploying Argon agents to facilitate C and C++ migrations to Rust. Another focused on memory profiling across multiple data centers, ultimately freeing over 300 tebibytes of memory. Additional savings were also identified through ongoing analysis of these systems.
Enhanced Capacity for Extended Technical Tasks
In its evaluations, Google reported a 77.9% score for Argon on DeepSWE v1.1, a benchmark for assessing extended software engineering performance. Results for finance, legal work, automation, and multimodal tasks have also been disclosed. Developed by Google DeepMind, Argon is part of the wider Gemini model family, which integrates coding tools with long-context reasoning and multimodal capabilities. Its expanded output capacity is tailored to support projects requiring multiple interconnected steps to achieve completion.
Cybersecurity remains a key focus during the initial deployment. Argon can identify, verify, and remediate software vulnerabilities within authorized defensive environments. Through its Scan for Good initiative, Wiz is utilizing the model to detect security flaws in public infrastructure. Google also reported a 68% score on CWE-bench v1, a benchmark dedicated to vulnerability remediation. Selected cybersecurity defenders are permitted to use Argon beyond standard security guardrails when working on approved security tasks.
Limited Public Availability During Phased Launch
No definitive date has been announced by Google for broad access to Gemini 4 Argon. The rollout is being conducted in phases, with feedback being collected from early users. Additionally, the company is participating in a voluntary U.S. government program that grants pre-release access to advanced AI models. The eventual rollout will include developers, enterprise clients, and consumers. Priority access is expected for paid API subscribers and Google AI Ultra members, although an exact launch date remains unspecified.
Google also clarified that Gemini 3.5 Pro will not be released. This model had initially been anticipated before the Gemini 4 series. Currently, Argon serves as the flagship model for complex reasoning and professional tasks. Other Gemini models continue to be available for users with varying needs in terms of performance and cost. For the time being, Gemini 4 Argon is limited to trusted testers, cybersecurity collaborators, and select early-access programs, with a wider release still pending.