Google’s Gemini 4 Argon is now open to developers. Here’s what it costs and who can use it
After a cautious, invite-only start, Google has made its most capable AI model widely available, with launch pricing that doubles on December 1.
Google has opened up Gemini 4 Argon, its newest and most powerful AI model, to paying developers, businesses and top-tier subscribers. The model became generally available on October 13, about two weeks after Google first revealed it and limited access to a small group of cybersecurity defenders.
For anyone who builds with AI, the big questions are simple: who can use it, what does it cost, and is it actually better? Here is what we know so far.
A slow, careful rollout
Google announced Argon on September 30, but it did not hand the model to everyone straight away. The first users were vetted security teams in a scheme Google calls the Fairwind Program. The idea was to give defenders a head start on a model that is very good at finding software weaknesses, before attackers could use it too.
Since October 13, that has changed. According to Google’s developer documentation, Argon is now available through:
- The Gemini API and Google AI Studio, for paid accounts (not the free tier)
- Google Cloud and Gemini for business customers
- The Gemini app and AI Mode in Search, for Ultra and Pro subscribers
- Google Antigravity, its AI coding tool, for AI Ultra subscribers
The Fairwind Program is also expanding to hundreds more partners, and security teams can still apply for a version with fewer cyber restrictions.
The price, and the December catch
Argon launched with introductory pricing of $2 per million input tokens and $10 per million output tokens. Cached input, meaning text the model has already seen, is heavily discounted.
That price does not last. Developer-focused reports based on Google’s pricing page say the introductory rate runs until November 30, 2026. From December 1, prices are due to double to $4 for input and $20 for output per million tokens. Teams testing Argon now should plan their budgets around the higher number, not the launch offer.
One more detail matters for cost: output pricing includes the model’s “thinking” tokens. A model that reasons at length before answering can use far more tokens than the visible reply suggests.
Is it actually better?
Google says Argon is aimed at three areas: software engineering, legal and financial knowledge work, and defensive cybersecurity. Google DeepMind’s Koray Kavukcuoglu has said it ranks first on the Vals Index, a benchmark that tests coding, finance and legal tasks.
The honest answer is that it depends on the task. In Google’s own published tables, Argon leads rivals on several tests but trails on others, including some coding and computer-use benchmarks. Company benchmark tables also compare against rivals’ self-reported numbers, which is not the same as a controlled, independent test. Early independent rankings place Argon among the top models, but not clearly ahead of every competitor.
The best benchmark is your own work. Run Argon on the tasks you actually do before you switch.
Why it matters
The launch shows how the biggest AI labs now release their strongest models: in stages, with security researchers first. As AI gets better at finding software bugs, companies are trying to make sure defenders benefit before attackers do. Expect this “defenders first” pattern to become normal.
It also shows that the price war in AI is not one-way. Launch discounts help a model gain users quickly, but businesses need to watch for price changes after the first few weeks.
What you should do
- If you are a developer: test Argon on a small set of real tasks before November 30 and compare the results and costs with your current model.
- If you run a business: budget using the December prices, not the launch offer.
- If you are a regular user: you will only see Argon in the Gemini app if you pay for a Pro or Ultra plan.
