Google unveils Gemini 4 Argon with advanced coding & cybersecurity capabilities, rolling out first to trusted cyber defenders

Google has officially unveiled Gemini 4 Argon, its new frontier AI model designed for complex software engineering, enterprise work and cybersecurity. The model is initially being offered to trusted cyber defenders as Google works on its safeguards before a wider rollout.

Gemini 4 Argon is built for longer, harder tasks

Google says Argon can handle long-horizon workflows across coding, legal and financial work, research and cybersecurity. One of its biggest upgrades is a 1 million-token output limit, up from 64,000 tokens in previous Gemini models, allowing the model to work through much longer tasks in a single run.

On DeepSWE v1.1, a benchmark for real-world software engineering tasks, Argon scores 77.9%, according to Google’s testing. Google also reports a 91.7% score on LVBench for long-video understanding and 51.3% on AutomationBench for end-to-end business workflows.

Google says Argon is already being used internally. Its agents have helped with large-scale code migrations and data-center memory optimization, with one project freeing more than 300 TiB of memory after deployment.

Image: Google

Cybersecurity is a major focus

Google is initially giving Argon to trusted cyber defenders through its Fairwind Program. The company says the model can autonomously find, validate and patch software vulnerabilities. On CWE-bench v1, Google reports Argon at 68%, tied for the top score in its comparison. Security company Wiz is already using Argon through its Scan for Good initiative, where Google says the model uncovered a critical vulnerability affecting healthcare software.

Google is also adding safeguards against prompt injection, misuse and model misalignment before expanding access. Argon will eventually reach developers, enterprises and consumers, starting with paid API customers and Google AI Ultra subscribers. A broader public release date has not yet been announced.