Google launches Gemini 4 Argon, but only cyber defenders get it first

Google announced a new top-tier AI model, Gemini 4 Argon, on 30 September (US time). It is pitched at deep reasoning, software engineering, knowledge work and cybersecurity defence. Most of us can't use it yet, but it's worth knowing about.

What's been announced

Argon is the first model in the Gemini 4 family. Google says it is built to keep reasoning through long, complex tasks. It also says its own staff already use it for daily work such as debugging and migrating code.

Access is limited at launch. Google is starting with a group of vetted security teams through its Fairwind Program. Those teams get a version without the usual cyber guardrails, so they can test the model's defensive abilities in full.

Google says wider access will follow, starting with paid API customers and Google AI Ultra subscribers. It hasn't given a date, and at the time of writing reports indicate the model isn't yet listed in the standard Gemini API.

Pricing

Google has published introductory API prices of US$2 per million input tokens and US$10 per million output tokens. Cached input, meaning text the model has already seen and stored, is discounted by 95%.

That is an introductory rate. Google's footnote says prices will rise to US$4 and US$20 per million tokens later, but it hasn't said when. If you plan to build on Argon, budget for the higher rate.

Why it matters for Australian businesses

• Nothing to switch on today. If you're already using Gemini, ChatGPT or Claude, there's no need to change tools this week. Wait until Argon is actually available to you.
• Security is the early focus. Google is putting this model's most powerful security features in defenders' hands first. That's a reminder that AI is also being used to find software weaknesses, so keep patching and review who has access to your systems.
• Costs will matter. If you or a developer are building automations on top of a model, the gap between introductory and later pricing can change the numbers. Caching can cut costs when you send the same background material repeatedly.
• Expect more competition. A new flagship from Google adds pressure on rivals to improve and price competitively. That usually works in customers' favour.

What to do now

Treat the benchmark claims as Google's own until independent testing catches up. When access opens, trial Argon on a small, real task. Good candidates are summarising a long contract, reviewing a spreadsheet process, or drafting a technical document. Compare the results and the cost with what you use now.

If you'd like help working out whether a new model is worth adopting in your business, that's the kind of question we help with at Tisso AI.