On September 30, 2026, Google announced Gemini 4 Argon on its public blog. The new AI model is available first to trusted cyber defenders in Google’s Fairwind Program. Google plans to offer it to developers, businesses, and consumers, but hasn’t given a public release date.
If you use ChatGPT, Claude, or Gemini every day, that access limit is probably the most frustrating part. Google has shared big claims and test scores, but you can’t switch on Argon in the Gemini app today.
What Happened
Google says Argon can help with coding, research, business tasks, and cybersecurity. In plain English, it can help write software, tackle detailed questions, and find and patch serious flaws. Google also claims Argon can produce up to 1 million output tokens, up from a 64,000-token limit. Tokens are small chunks of text that AI systems count as they read and write.
That’s room for a very long answer or work product. It doesn’t mean you’ll be able to ask the Gemini app for a million-token reply. Google hasn’t explained how that limit would work in a consumer app, and the claim hasn’t been independently tested here.
Who Can Use It Now?
Google says initial access is for trusted cyber defenders in the Fairwind Program. It says Argon can find, check, and patch critical software flaws. That helps explain why Google is starting with a selected group.
Google says broader access will begin with paid API customers and Google AI Ultra subscribers, after the initial cyber-defender testing. The announcement gives no general-availability date. A plan isn’t a release date (annoying, I know).
What Pricing Has Google Announced?
Google has announced API pricing for developers and businesses that use the model through software. These aren’t Gemini app subscription prices. The announcement also doesn’t mean a public Argon API is available today.
| Announced API rate | Input | Output |
|---|---|---|
| Introductory pricing | $2 per 1 million tokens | $10 per 1 million tokens |
| Later pricing | $4 per 1 million tokens | $20 per 1 million tokens |
Google says cached input tokens get a 95% discount under its announced pricing. It hasn’t said when the later rates begin. If you use Gemini on your phone or Windows PC, these figures aren’t a price for adding Argon.
How It Stacks Up
Google reports that Argon leads the listed rivals on the three tests selected below. Results vary across the full chart: on FrontierSWE v2, Argon scores 55.0% versus GPT-6 Astra’s 65.5%. The scores below are a small selection from the comparison chart in Google’s Argon announcement, focused on knowledge work and coding. They are Google’s results, not our own testing or the full benchmark set.
| Benchmark | Gemini 4 Argon | GPT-6 Astra | Claude Fable 5.1 | Claude Opus 5.5 |
|---|---|---|---|---|
| Vals Index Knowledge Work | 68.9% | 63.1% | 65.8% | 67.0% |
| AutomationBench Knowledge Work | 51.3% | 41.4% | 31.4% | 42.5% |
| DeepSWE v1.1 | 77.9% | 74.1% | 67.4% | 74.2% |
Source: Google’s published Gemini 4 Argon comparisons. These are vendor-reported figures; we have not independently tested Argon.
Vals Index weights work across finance, coding, legal and tax domains; AutomationBench measures end-to-end business-task execution. DeepSWE tests software engineering. If independent tests back up Google’s scores, Argon could be useful for people who work with code or dense documents.
But a benchmark is like a driving test: it can’t tell you what your daily commute will feel like. These scores don’t show whether Argon will write better emails or help more with your spreadsheets. They also don’t settle which AI assistant is best overall.
Google chose GPT-6 Astra, Claude Fable 5.1, and Claude Opus 5.5 for this table. Those results don’t tell us how Argon compares with the latest public options or with a public Gemini model on the same tasks. For now, it’s Google’s comparison, not a tested guide to choosing an AI subscription.
Google also calls Argon multimodal, meaning it can work with more than one kind of information. That could help if your work mixes text with other material. Google hasn’t said which inputs a future consumer version will accept. Its cybersecurity claims explain the Fairwind rollout, though independent hands-on tests are still needed to judge how well Argon fixes real flaws.
Early Reaction
Public reaction is still taking shape, and this article does not claim a broad consensus.
It’s a fair question, though: how much should a score sway you when you can’t try the model? Google’s numbers make Argon worth watching. They don’t show what using it alongside ChatGPT, Claude, or Gemini will feel like.
Our Take
Verdict: I’d stick with the AI assistant you already use for now. Argon has limited availability, its scores need independent testing, and Google hasn’t set a public release date. If you’re an eligible Fairwind cyber defender, the model is relevant to your work. Developers and businesses can note the API rates and revisit them when access and practical limits are clear.
What this means on Windows, Android, and the web
Google’s announcement describes restricted initial access and a planned expansion, not a general consumer launch. Keep using the tools available in your account; check the official announcement before changing subscriptions or expecting an Argon option in your app.