Google Unveils Gemini 4 Argon Frontier Model
Google has unveiled Gemini 4 Argon, its new frontier AI model for complex, long-running workloads. The model targets software engineering, enterprise knowledge work, and cybersecurity.
However, most users cannot access Argon yet. Google is initially providing it to selected cybersecurity experts through its Fairwind program. Meanwhile, Google says it plans broader access for developers, enterprises, and consumers after further testing and safeguards.
Gemini 4 Argon Targets Complex Workflows
Argon focuses on tasks that require sustained reasoning across multiple steps. In particular, Google highlights coding, legal and financial work, research, and cybersecurity defence.
The model also supports a much larger output limit. Argon can generate up to 1 million tokens, compared with 64,000 tokens in earlier Gemini models. Therefore, it can handle substantially longer tasks within a single workflow.
Google says its teams already use Argon for specialised coding, research, and writing. Additionally, the company says Argon has helped optimise internal systems and support large-scale software engineering work.
The company also published benchmark results showing strong performance. For example, Google reports a 77.9% score on DeepSWE v1.1, a software-engineering benchmark.
Cybersecurity Leads the Initial Rollout
Cybersecurity remains a central focus of Argon’s initial deployment. Google says the model can identify, validate, and patch critical software vulnerabilities. Consequently, the company is starting with trusted cybersecurity defenders before expanding access.
Google has also introduced additional safeguards for the model. These systems monitor its reasoning and can intervene when its behaviour moves outside defined boundaries. The company says this phased approach reflects the capabilities and risks associated with frontier models.
Moreover, Google is participating in the US government’s voluntary pre-release access program. This gives selected government agencies access to advanced models before they are widely available.
Pricing and Broader Availability
Google has announced introductory API pricing for Gemini 4 Argon. The promotional rate is $2 per million input tokens and $10 per million output tokens. Cached input tokens receive an additional discount.
However, Google has not provided a specific date for general availability. The company says paid API customers and Google AI Ultra subscribers will receive access after the initial cybersecurity phase.
Argon also marks Google’s return to the top end of frontier model development after months focused on smaller Gemini Flash releases. Meanwhile, competitors, including OpenAI and Anthropic, have continued to expand their advanced models.
