Summary

Google says Gemini 4 Argon is designed for long, complex software, enterprise and cybersecurity tasks. It is initially rolling out to trusted cyber defenders, with broader access planned.

Google announced Gemini 4 Argon on September 30, 2026, and says it is beginning a limited rollout through its Fairwind Program to trusted cyber defenders. The company describes Argon as a model for extended, multi-step work in software engineering, legal and financial knowledge work, and cybersecurity. Google says it plans to broaden access to developers, businesses and consumers after gathering feedback from early testers and refining safeguards.

Contents

A phased release focused on cyber defence

Google says Argon can autonomously find, validate and patch critical software vulnerabilities. The company says it will provide the model without cyber guardrails to trusted defenders and its own internal teams, while it gradually expands access. Google also says it is participating in the U.S. government’s voluntary process for pre-release model access.

The staged rollout reflects the model’s security use: vulnerability discovery and patching can support defenders, while access is initially limited to a selected group. Google says its early testers will help it iterate on safeguards before a wider release. The company has announced introductory prices of $2 per million input tokens and $10 per million output tokens; cached input tokens are priced at 95% off the input-token price.

Longer outputs and task performance

Argon’s output-token limit is 1 million, up from 64,000 in the previous limit Google cites. Output tokens are the units a model generates in its response; a higher limit allows it to produce much longer outputs in one run. Google says this capacity gives Argon room to sustain reasoning through more complex workflows and generate hundreds of thousands of tokens in a single trajectory.

Google reports a 77.9% result on DeepSWE v1.1, an evaluation of long-horizon software-engineering tasks. It also reports a score of 51.3% on AutomationBench, which measures end-to-end execution across business functions, and 91.7% on LVBench, an evaluation of long-video understanding. The company describes Argon as the leading model on the Vals Index, which assesses work across finance, coding, legal and tax tasks.

These results indicate the kinds of tasks Google is targeting: not just short question-and-answer exchanges, but work that involves multiple steps, tools or information sources. The figures are Google’s reported benchmark results, each tied to a particular evaluation.

Internal engineering and security examples

Google says its teams are using Argon for coding, research and engineering work. One example involves libgav1, Google’s open-source video decoder: the company says Argon agents replaced 32,000 lines of SIMD code in an existing Rust port with safe Rust that the compiler could automatically vectorise. Google reports that the resulting decoder runs 2.7 times faster than that Rust port and produces identical video output.

For cybersecurity, Google says cloud-security company Wiz is using Argon in its Scan for Good initiative, which aims to find and remediate high-risk exposures in critical public infrastructure at no charge. In an early demonstration, Google says Argon identified a critical vulnerability exposing sensitive personal information in healthcare software used by hospitals worldwide. It also reports that Argon tied for first place on CWE-bench v1, a vulnerability-remediation evaluation, with a score of 68%.

The initial release therefore pairs a model Google says can perform advanced defensive security work with access limited to trusted defenders. Wider availability is planned, but the announcement does not give a release date.

Sources