Google limits Gemini 4 Argon to trusted testers as cyber testing begins
Google says the model can handle coding, knowledge work and cybersecurity tasks, but has given no general release date or API pricing.
Google has announced Gemini 4 Argon, a new AI model available initially to a small group of trusted testers, including cyber defenders. Wider access is planned after cybersecurity testing, but the company has not set a release date.
Google says Argon performs strongly in coding, knowledge work and cybersecurity. The company reports a 77.9% score on the DeepSWE v1.1 software engineering benchmark, a result not independently verified in the supplied material.
The company also says Argon supports outputs of up to one million tokens and has been used internally to save memory in its data centres and migrate code from C and C++ to Rust.
Google says partner Wiz used the model to identify a critical vulnerability in a system used by hospitals worldwide. It has not provided details of the vulnerability or how it could affect the system.
Google says paid API users and Google AI Ultra subscribers will be among the first to receive broader access after the cyber testing. API pricing remains unannounced.
