Google announced Gemini 4 Argon on September 30, 2026, its newest frontier model, and for now you cannot pick it in the Gemini app. Google says it is rolling out first to Google AI Ultra subscribers and paid API customers, while vetted cyber defenders already have access through a program called Fairwind.
If you use Gemini on a Pixel or an iPhone, the real question is when Gemini 4 Argon reaches your phone. Google has only said “soon.”
TL;DR: Gemini 4 Argon is Google’s new top model, built for coding, long workflows and cybersecurity defense. It starts with AI Ultra subscribers and paid API customers, with no consumer date given. API pricing doubles after an introductory period, and the 1M token output limit is the biggest practical change.
What Gemini 4 Argon is built to do
Google describes Gemini 4 Argon as a model for deep reasoning across complex, long-horizon workflows, with training focused on coding, knowledge work, cybersecurity defense and creative writing.
The company published a set of benchmark results alongside the announcement. They are Google’s own numbers, so treat them as a starting point until independent testing arrives.
| Benchmark | What it measures | Gemini 4 Argon result |
|---|---|---|
| DeepSWE v1.1 | Software engineering tasks | 77.9% |
| CWE-bench v1 | Finding security vulnerabilities | 68%, tied for first |
| AutomationBench | Workflow automation (Zapier) | 51.3%, ranked first |
| LVBench | Long video understanding | 91.7% |
A benchmark lead tells you what a model can do in a lab. It says little about whether the Gemini app on your phone will feel any different.
Who gets Gemini 4 Argon first
Google AI Ultra subscribers and paid API customers come first, according to the announcement. Developers, enterprises and consumers are all listed for a wider rollout, but Google gave no dates for any of them.
Google also says it is taking part in the U.S. government’s voluntary pre-release model access process, with access expanding gradually. If you are on the free Gemini app, expect to wait.
That staged approach also explains why the pricing arrived before the app did.
Gemini 4 Argon API pricing and the 1M token output limit
Gemini 4 Argon costs $2 per million input tokens and $10 per million output tokens during an introductory period. After that it moves to $4 and $20, and cached input tokens get a 95% discount off the input price.
The output limit is now 1 million tokens, up from 64,000. For a developer, that is the difference between a model that has to stop and be restarted halfway through a large code migration and one that can keep writing.
Google points to its own use: it says Argon scaled C and C++ to Rust migrations up to 800,000 or more lines for the Fuchsia Zircon kernel, and produced a memory safe video decoder that runs 2.7 times faster than the Rust port. Those are Google’s claims about its own code, and the output limit is what makes jobs that size plausible.
The Fairwind Program and the cybersecurity angle
Fairwind gives cyber defenders what Google calls frontier-level cybersecurity defense capabilities, without the guardrails the public model keeps. Google says the goal is finding and fixing vulnerabilities in critical infrastructure, and it names Wiz, which is using Argon for a “Scan for Good” effort on healthcare software.
The public model is a different story. Google says it is designed to refuse harmful cyber and CBRN attack requests while still allowing legitimate dual-use research, and that it is its most resilient model yet against prompt injection.
That split, a looser model for vetted defenders and a restricted one for everyone else, is the detail most coverage will skip. It also sets the tone for how Google plans to ship powerful models from here.
What to watch before Gemini 4 Argon reaches your phone
Nothing in Google’s announcement says Gemini 4 Argon is coming to Android or iOS on a set date. The places most likely to feel it first are the Gemini app for Windows and Gemini in Chrome, but that is an inference, not something Google has said.
The practical advice is simple. If you pay for AI Ultra, check the model picker in the coming weeks. If you do not, the benchmark chart is interesting but changes nothing on your home screen yet.
