Google DeepMind introduced Gemini 4 Argon, a new AI model that handles complex, long-horizon workflows. It is designed for tasks like coding, legal work, finance, and cybersecurity defense.
The model features an industry-leading 1 million token limit, allowing it to generate and reason over very long sequences. This enables it to solve difficult problems in a single run.
Gemini 4 Argon is already used internally at Google. It helps with tasks like optimizing quantum algorithms, analyzing telemetry data, and migrating large codebases. For example, it improved quantum subroutines by 40% in minutes.
It also saved over 300 terabytes of memory across Google’s data centers by analyzing telemetry data and applying optimizations. This can lead to total savings of up to 1 petabyte.
In software engineering, Argon set a new state of the art with a score of 77.9% on the DeepSWE benchmark. It also performs well in finance, legal research, and automation tasks.
The model is trained to understand visual data, such as analyzing charts and videos. It scored 91.7% on the LVBench video understanding benchmark.
Google is rolling out Argon to trusted cybersecurity defenders through its Fairwind Program. It can find, validate, and patch software vulnerabilities autonomously.
Safety and testing are priorities before wider release. Google plans to expand access gradually and gather feedback to improve guardrails.
Source: https://deepmind.google/blog/gemini-4-argon-our-next-era-of-frontier-intelligence/