Google launches Gemini 4 Argon, says it excels at cybersecurity

Alphabet has launched Gemini 4 Argon, a new AI model built to handle tasks such as coding, research, and writing — though Google says cybersecurity is where it has a particular edge.

Argon is being rolled out only to a select group of the company’s cyber partners through its Fairwind Program, Google’s security initiative. The model was trained specifically for defensive cyber work, and Google says it can autonomously find, validate, and patch critical software vulnerabilities.

The company also says Argon is strong at coding and engineering, noting that its own staff have already been using the model for daily work, including debugging and codebase migrations. Google further touts Argon’s ability to parse visuals, such as analyzing the contents of long videos or charts. “Built to sustain deep reasoning across complex, long-horizon workflows, Argon is fundamentally changing the way we work and build at Google,” the company said in a blog post Wednesday.

Argon arrives as top AI labs rush to release increasingly powerful models, competing to outdo each other even as those same firms warn that AI could spin out of control. OpenAI recently released Astra, which it hailed as its best model yet, and Anthropic’s Fable was released earlier this year with similar rhetoric. Google claims Argon scored significantly higher than OpenAI’s GPT-6 Astra and Anthropic’s Fable and Opus models across a variety of benchmarks, citing the increasingly popular AI benchmarking startup Vals to show Argon is currently the leading model on its AI model index.

Google, once seen as trailing in the AI race, has recently found success with Gemini: the company said in August that its app had over a billion monthly users, a figure that puts it on par with OpenAI, which has also recently said ChatGPT reached a billion monthly users.

Why it matters

Framing Argon around autonomous vulnerability-patching — rather than general chat or reasoning benchmarks alone — signals Google is trying to carve out defensive cybersecurity as a distinct competitive front in the AI model race, separate from the raw-capability contest with OpenAI and Anthropic.