Google has announced Gemini 4 Argon. It says the model can work on hard coding and security jobs that take many steps. That sounds exciting. The catch is simple: you cannot just open the Gemini app and try it today. Google is first giving access to trusted cyber defenders in its Fairwind Program.
What is different?
Google says Argon can produce up to one million output tokens in one task. A token is a small piece of text. This is a limit on what the model can write out, not a claim that it can read one million tokens of source material at once. More room to write can help with a long task; it does not prove that every answer is right.
Google reports using Argon for software work, research, and finding and fixing security flaws. These are the company's claims. WiredBench has not tested the model.
Why is access so limited?
A tool that finds software weaknesses could help defenders, but it could also be misused. Google says it wants feedback and stronger safeguards before a wider release. The announcement gives no firm date for public access.
So if you are wondering whether to switch tools this week, there is nothing to switch to yet. Watch for a clear launch date, who gets access, and independent tests. Even when the model becomes available, a human should check any code or advice it produces.
Image note: The lead image is an AI-generated editorial illustration, not a photograph of Google staff or Gemini 4 Argon in use.
