Google Unveils Gemini 4 Argon, But You Can’t Use It Yet

Google has unveiled Gemini 4 Argon, its latest frontier AI model, with a focus on complex software engineering, enterprise knowledge work and cybersecurity. But unlike a typical Gemini launch, most developers, businesses and consumers can’t actually use it yet. Instead, Google is starting with a restricted rollout to selected cybersecurity experts through its Fairwind Program. 

The company says it will gradually expand access while gathering feedback and strengthening its safety measures, although it hasn’t said when Argon will become generally available. The unusual release means some of Argon’s most advanced capabilities are already being used in the real world, while everyone else will have to wait.

What Is Gemini 4 Argon?

Gemini 4 Argon is Google’s new top-tier frontier model, meaning it represents the most advanced end of the company’s current AI development. Google has designed it to handle complex tasks that can run across many steps, rather than simply responding to a single prompt or completing a short piece of work. 

That includes software engineering and enterprise work in areas such as finance and law. Google has also placed a particular focus on cybersecurity, training Argon to autonomously find, confirm and fix critical software vulnerabilities. To support longer tasks, Google has increased Argon’s maximum output from 64,000 tokens to one million. 

Tokens are the small pieces of text an AI model processes and generates, so the larger limit allows Argon to produce and work through much longer outputs without having to stop and restart. Google says thousands of its employees are already using the model for specialised coding, research and writing. 

Its engineers have also used Argon for work ranging from everyday debugging to large code migrations and algorithm design. The company’s own benchmark results put Argon ahead of GPT-6 Astra and Claude Opus 5.5 on several enterprise and coding tests, although it doesn’t lead across every measure. 

On FrontierSWE v2, for example, Argon scored 55%, compared with 65.5% for GPT-6 Astra and 62.3% for Claude Opus 5.5.

Why Is Google Restricting Access To Gemini 4 Argon?

For now, Google is giving Argon to a select group of cybersecurity defenders through Fairwind, its programme for giving trusted security experts access to advanced AI capabilities. Cybersecurity is also one of the reasons Google is taking a more cautious approach to the wider rollout. 

The same AI capabilities that can help defenders discover and repair software vulnerabilities could potentially be misused to help carry out cyberattacks. Google says it’s strengthening Argon’s safeguards against cyber and chemical, biological, radiological and nuclear misuse before making the model broadly available. 

It’s also working on protections against prompt injection attacks, where malicious instructions are used to manipulate how an AI system behaves. Trusted cyber defenders and Google’s own internal teams are getting a different version of that experience. Google says it will give them Argon without its normal cybersecurity restrictions so they can use the model’s full defensive capabilities. 

That work has already started. Google-owned security company Wiz is using Argon through its Scan for Good initiative, which looks for vulnerabilities in critical public infrastructure. Google says the model has already found a critical flaw exposing sensitive personal information in healthcare software used by hospitals around the world.

Gemini 4 Argon Replaces Google’s Delayed Gemini 3.5 Pro

Argon’s arrival also marks a change in Google’s model roadmap. Google had previously planned to release Gemini 3.5 Pro in June. The model never arrived, and Reuters reports that Google has now abandoned those plans and moved directly to Gemini 4 following months of delays. 

Reuters reports that internal changes at Google DeepMind and leadership departures contributed to delays in the company’s AI development plans. During that time, competition at the frontier continued to intensify as OpenAI and Anthropic released their own advanced models. 

Gemini 4 Argon now gives Google another model competing at that level, but the company isn’t rushing to put its full capabilities into everyone’s hands.

When Will Gemini 4 Argon Be Available?

Google hasn’t announced a general release date for Gemini 4 Argon. The company says it will continue collecting feedback from its initial group of cybersecurity experts and trusted evaluators while refining its safeguards. It’s also participating in the US government’s voluntary process for providing pre-release access to frontier AI models. 

Once the wider rollout begins, Google says paid API customers and Google AI Ultra subscribers will be among the first to receive access. Argon will initially cost $2 per million input tokens and $10 per million output tokens, before rising to $4 and $20 respectively after the introductory period. 

For everyone outside Google and its selected security partners, however, Gemini 4 Argon remains something they can read about, but not yet use.