Google releases Gemini 4 Argon AI model with phased rollout

In short: Google has released Gemini 4 “Argon,” its newest top-end AI model, and it is sharing it first with a limited set of companies and governments.

What happened

Google says Gemini 4 Argon matches or beats leading rivals like OpenAI and Anthropic on tests for coding and cyber security. The company is positioning it for business work in areas like software, finance, law, and tax.

This launch follows a delay. Google had previously suggested it would release an earlier model, described as Pro 3.5, but people familiar with the process told the Financial Times that Google decided not to ship it because it was taking too long and no longer looked strong versus competitors.

Google says Argon can handle harder, longer tasks partly because it has a much larger “context window.” This is the amount of information the model can keep in mind at once, like having a bigger desk to spread out documents while you work. Google said this capacity increased 16-fold.

Google is distributing Argon through a program called “Fairwind” to a select group. The goal is to let them find and report cyber weaknesses before a wider release. Google also said Argon is designed to refuse harmful requests related to things like cyber attacks and CBRN threats (chemical, biological, radiological, and nuclear).

Why it matters

Many companies are looking for AI tools that can safely help write code and spot security problems, but they also worry about cost and misuse. Google did not share pricing details, so businesses will be watching how expensive Argon is and how well its safety checks work in real-world use.

Source: Financial Times