Google has introduced Gemini 4 Argon, its latest frontier AI model designed to handle complex tasks that require long-running reasoning. The company says Argon delivers strong performance in software engineering, enterprise work, research and cybersecurity. Google is initially making the model available to a limited group of trusted cyber defenders through its Fairwind Program. The company is taking a phased approach to the rollout to improve the model’s safety measures. According to Google, Gemini 4 Argon outperforms GPT 6 Astra on several benchmarks.
One of the biggest changes with Gemini 4 Argon is its expanded output limit. The model can generate up to 1 million tokens, compared with 64,000 tokens in the previous generation. This allows Argon to work through longer and more complicated tasks in a single session.
Today we’re introducing Gemini 4 Argon.
— Google (@Google) September 30, 2026
It delivers frontier performance in complex workflows across real-world software engineering, knowledge work, and cybersecurity defense with an industry-leading 1M token output limit. pic.twitter.com/oDaCoOLu5E
The company said Argon is already being used internally by thousands of Google employees. “Argon is helping our quantum computing researchers optimise the spacetime resources (qubits × gates) of subroutines that bottleneck important applications. In one example, it beat the published baseline by 40% in a matter of minutes,” Google revealed.
The tech giant also claimed that the Gemini 4 Argon AI model is being used for large-scale code migrations.
According to Google, Argon achieved 77.9 per cent on DeepSWE v1.1, a benchmark for real-world software engineering tasks. It also claims the model ranks first on AutomationBench with a score of 51.3 per cent.
Cybersecurity is another major focus. Argon can reportedly find, verify and fix software vulnerabilities.
“Argon is also our most resilient model yet against indirect prompt injections, where malicious instructions or context are used to hijack a model’s behavior,” the company said.
Google is taking a gradual approach for the Gemini 4 Argon rollout. The model is currently available to selected cyber defenders and trusted testers through the Fairwind Program.
The company says it is also participating in the US government’s voluntary process for giving authorities access to frontier models before release. Feedback from early users will be used to improve Argon’s safeguards before a wider launch.
Google plans to make Argon available to developers, businesses and consumers. The rollout will begin with paid API customers and Google AI Ultra subscribers.
The introductory price is set at $2 per million input tokens and $10 per million output tokens. Cached input tokens will cost 95 per cent less than the standard input price.