Gemini 4 Argon: What Google's New AI Model Is and Who Can Use It
Google opened Gemini 4 Argon to subscribers and developers on October 13, 2026. Who gets it, what it costs, how it fits into this fall's AI race, and why its benchmarks need a second look.
CAESVRA JournalOctober 16, 20262 min read
Google calls Gemini 4 Argon its most powerful model yet. Since October 13, 2026, many more people can try it. Here is what it is, who gets it and what it costs, without the hype.
The short version
- What: Gemini 4 Argon, Google's newest and largest AI model.
- Announced: September 30, 2026, first only for cybersecurity partners.
- Opened up: October 13, 2026, for paying subscribers and developers.
- Free? No. Free users and the AI Plus plan do not get it.
- Developer price: $2 per million input tokens and $10 per million output tokens until November 30, then $4 and $20.
Who can use Gemini 4 Argon
| Who | Access |
|---|---|
| Google AI Pro and Ultra subscribers | yes, in the Gemini app and in AI Mode in Search |
| Developers | yes, through the paid Gemini API and AI Studio |
| Antigravity (Google's coding tool) | Ultra since October 13, Pro since around October 15 |
| Free users and AI Plus | no |
When Google first announced the model on September 30, only partners in its Fairwind cybersecurity program had access. The broad rollout came two weeks later, in a post by Koray Kavukcuoglu on Google's blog.
How to use it
With a Google AI Pro or Ultra subscription, Gemini 4 Argon is available in the Gemini app and in AI Mode in Google Search. Developers can use it in AI Studio or through the paid Gemini API. If you build on it, plan for the price change on November 30.
How good is it?
Google says Gemini 4 Argon ranks first on the Vals Index and beats OpenAI's GPT-6 Astra and Anthropic's current models. Those are Google's own claims. CNBC pointed to a Bloomberg report that questioned its real-world coding performance, which Google disputes. Independent tests will take a few more weeks. Until then, treat the benchmark charts as marketing.
The rest of the AI race this fall
Gemini 4 is not the only story:
- OpenAI cancelled GPT-6.1 Astra at the end of September after the model failed internal safety tests. The Wall Street Journal broke the story and OpenAI confirmed it. It was not delayed, it was scrapped.
- The US Federal Trade Commission demanded records from OpenAI and Anthropic about safety incidents on October 14.
- SpaceX is reportedly in talks to raise about $40 billion in debt to buy Nvidia AI chips. That is a report about talks, not a finished deal.
Before and after
AI moves in versions, and every version claims to be the cut. Most are not. What makes this fall different is not one model, but that the companies themselves are starting to stop, cancel and get questioned. That may turn out to be the bigger before and after.
From another corner of tech this week: Apple's first foldable, the iPhone Duo.
Sources
→More from the journal
All entries →
