CALIFORNIA / RankWire.AI / – Google continues to push the boundaries of artificial intelligence development with the launch of Gemini 4 Argon, a new flagship model designed for sophisticated professional applications. The company unveiled Argon on Sept. 30, positioning it as the top model in the Gemini 4 series. It is capable of supporting intricate software development, financial analysis, legal research, and cybersecurity tasks. Additionally, the model excels at processing lengthy sequences of reasoning and execution. Currently, access remains limited, with select cybersecurity professionals engaging with Argon via the Fairwind Program.
One of Argon’s key enhancements is the expanded maximum output limit, now reaching 1 million tokens—a significant jump from the previous cap of 64,000 tokens. This increase enables the model to handle extended tasks without needing to split work across multiple sessions. The introductory API pricing begins at $2 per million input tokens, with output tokens priced at $10 per million during the same period. Cached inputs are offered at a 95% discount. Future pricing is expected to increase to $4 for input tokens and $20 for output tokens.
Currently, thousands of employees within Google are utilizing Argon for coding, research, and various writing assignments. Internal teams have also tested the model on projects such as data center optimization and large-scale software migrations. For example, one initiative involved using Argon agents to facilitate the transition of C and C++ codebases to Rust, while another focused on memory profiling across data centers. The optimization efforts resulted in freeing over 300 tebibytes of memory, with teams discovering further savings through ongoing analysis of the same systems.
Enhanced support for lengthy technical tasks
According to Google, Argon scored 77.9% on DeepSWE v1.1, a benchmark assessing extended software engineering performance. The company also published results related to finance, legal research, automation, and multimodal tasks. Developed by Google DeepMind as part of the broader Gemini model family, Argon combines coding capabilities with long-context reasoning and multimodal functions. Its increased output capacity is tailored to support workflows involving multiple interconnected steps before reaching completion.
Cybersecurity remains a core focus during this initial phase. Argon can detect, verify, and patch software vulnerabilities within approved security 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 centered on vulnerability remediation. Select defenders can also utilize Argon without standard cyber guardrails when working on authorized defensive security operations.
Limited public access amid phased deployment
Google has not announced a definitive date for widespread public availability of Gemini 4 Argon. The company is adopting a phased release strategy, actively collecting feedback from early users. It also participates in a voluntary U.S. government process that grants pre-release access to advanced AI models. In the future, access will extend to developers, enterprise clients, and consumers. Priority access is expected for paid API subscribers and Google AI Ultra members, though an exact launch date has not been set.
Google also confirmed that Gemini 3.5 Pro will not be released. This model was initially anticipated prior to the Gemini 4 series. Instead, Argon now represents the latest flagship offering for high-demand reasoning and professional workloads. Other Gemini models remain available to meet different performance and pricing needs. For the time being, Gemini 4 Argon is primarily accessible to trusted testers, cybersecurity partners, and early-access programs, with broader rollout still pending.