Gemini 4 Argon is here, but cyber defenders get it first
Google's new frontier model, Gemini 4 Argon, is priced from $2/$10 per million tokens. It goes to vetted cyber defenders first. What's confirmed.
Stated by the maker, publisher, or platform holder.
Six days after DeepMind's new head said Gemini 4 would come "as soon as possible," Google announced it. Gemini 4 Argon went official on 30 September, and Google calls it a new "frontier model."
The speed fits what Koray Kavukcuoglu said on stage at The Information's AI Agenda Live summit on 24 September, his first public appearance since taking over Google DeepMind in August: that Gemini 4 was in post-training, was already running internally to power Google's Antigravity coding tool, and that he hoped to release it "much earlier" than the end of the year.
There's a catch. You can't use it yet unless you're one of a small group of cybersecurity defenders. Everything below is from Google's own announcement.
What's confirmed
Who actually gets it
There's no public date. That's the same approach the big labs took with their most capable security models earlier this month, which we covered in the cyber-AI arms race: the more a model can find and fix vulnerabilities, the more carefully it's released.
How it compares
Google says Argon beats OpenAI's GPT-6 Astra and Anthropic's latest models on several benchmarks. Those are Google's own comparisons, on tests Google chose. Independent results will come once more people can use it.
The timing is pointed. It lands a day after OpenAI's DevDay, where GPT-6.1 Sol offered near-top performance at a fifth of the price. Argon's introductory price of $2 and $10 per million tokens puts it right in that fight.
Our read
Gemini 4 is real, and on Google's numbers it's at or near the top. For most people, though, nothing changes today. The release to cyber defenders first is the story: Google is treating its best model's security skills as something to hand out carefully. We'll update this when it reaches the API and AI Ultra.