Google Unveils Gemini 4 Argon AI Model for Coding and Cybersecurity

Google has unveiled Gemini 4 Argon, a next-generation artificial intelligence model specialized for coding, professional work, and cybersecurity. Internal evaluations of its practical application have been divided.
Google described Gemini 4 Argon as an AI model designed specifically for software development, legal and financial expertise tasks, and cybersecurity defense. In internal benchmarks, the model achieved a score of 77.9% on the DeepSWE v1.1 software engineering benchmark, with output limits increased from 64,000 tokens to one million tokens.
Google reported that internal engineers have begun using Gemini 4 Argon for tasks such as codebase migration and data center memory optimization. One example cited involved an agent analyzing data center profiling data to save over 300TiB of memory, while another case saw its use in migrating more than 800,000 lines of Fuchsia OS code.
However, Bloomberg reported that internal Google staff have differing opinions on the model's benchmark results versus real-world performance. Some employees noted difficulties with coding tasks, while others agreed the model represents cutting-edge capabilities across multiple fields. Gemini 4 Argon is currently available only to trusted cybersecurity partners on a limited basis, with no set date for general public release.
Korean Source
This article is an English localization of a Korean-language crypto news report. Original headline: 구글, Gemini 4 Argon 공개…내부 실무 평가는 엇갈려