Google DeepMind announces Gemini 4 Argon frontier model
Google DeepMind has announced Gemini 4 Argon, a frontier model designed for extended reasoning across software engineering, enterprise work and cybersecurity. The model is initially being provided to trusted cyber defenders through the Fairwind Program, while the company continues a phased safety process and gathers feedback before wider access.
DeepMind says Argon is already used in internal Google workflows, including coding, research and writing. Reported examples include beating a published quantum-optimization baseline by 40%, freeing more than 300 TiB of data-center memory after rollout, and helping migrate large C/C++ codebases to Rust. In one libgav1 project, an Argon-generated Rust decoder ran 2.7 times faster than an earlier Rust port with identical video output.
The model’s output limit has increased from 64K to 1M tokens. DeepMind reports leading results on several evaluations, including 77.9% on DeepSWE v1.1, 51.3% on AutomationBench and 91.7% on LVBench. It also says Argon can identify and patch vulnerabilities, and ties for first on CWE-bench v1 with 68%. The planned introductory price is $2 per million input tokens and $10 per million output tokens. Broader availability will begin with paid API customers and Google AI Ultra subscribers.
Written by Sakhanda Wire AI from reporting by DeepMind. Facts, quotes and full context are in the original article below.