Announced, Not Available
Google announced Gemini 4 Argon on 30 September 2026 and said it is rolling out first to a set of trusted cyber defenders through its Fairwind Program. The company calls its release phased and says it is taking part in the US government's voluntary process for pre-release model access while it gradually expands access.
General availability will start with paid API customers and Google AI Ultra subscribers once testing is done. Ars Technica notes that Google made no specific timeline promise, so for most buyers this is a price list and a benchmark sheet, not a product.
What Google Says It Can Do
| Claim | Google's figure |
|---|---|
| DeepSWE v1.1, long-horizon software engineering | 77.9 percent, a new state of the art |
| CWE-bench v1, vulnerability remediation | 68 percent, tied for first |
| AutomationBench, business functions | 51.3 percent, ranked first |
| LVBench, long video understanding | 91.7 percent |
| Output limit | 1 million tokens, up from 64,000 |
Google also reports internal results: agents that freed more than 300 TiB of memory across its data centres, with an estimated 500 TiB to 1 PiB in total, and C and C++ to Rust migrations that scale up to more than 800,000 lines for the Fuchsia Zircon kernel. Wiz, a partner, says the model found a critical flaw in healthcare software used by hospitals worldwide, without naming the software.
Treat all of this as vendor and partner reporting until independent testers have run the model.
The Price Is A Promotion
Google set an introductory price of 2 dollars per million input tokens and 10 dollars per million output tokens, with cached input at 95 percent off. After the introductory period the price becomes 4 and 20 dollars. Anthropic's Sonnet 5.5, released two days earlier, costs 2 and 10 dollars, so the standing Argon price is twice that of a rival that is already on sale.
The output limit changes the arithmetic. A single response that uses the full 1 million output tokens would cost 10 dollars at the introductory rate and 20 dollars after it. Google says the long limit lets the model finish hard problems in one trajectory, which is useful and also a reason to cap output per request.
Google also says it monitors the model's chain of thought and actions and can stop execution when the model steps out of bounds. For anyone running agents, that is a governance feature to ask other vendors about.
What To Do Before It Ships
Do not plan a migration around a model that has no release date. Budget at the 4 and 20 dollar rate, because the introductory price is a promotion, and keep your current vendor as the default until Argon is on sale and priced.
When access opens, run your own tasks and measure cost per finished job, including the cost of long outputs. Cap output tokens per request before you let any agent loose on a model that can write a million of them.
If your team does defensive security work, ask Google how the Fairwind Program decides who gets the version without cyber guardrails, and plan for access to be restricted for some time.
Update, 1 October 2026: Bloomberg reported that some Google employees found Gemini 4 does less well in real work than its benchmarks suggest and struggles with certain coding tasks. Google said it would be inaccurate to say Gemini 4 underperforms in coding.
Servola Journal
We do this for everyone trying to keep up with what technology is doing to our lives. The people who build it, and the people it happens to. The Servola Journal exists so that what we learn belongs to all of them.
Nobody pays us for this. No ads, no paywall, free to everyone. We just believe that understanding what's happening to all of us shouldn't depend on who can afford to pay for it.
If it gave you something today, tell us to keep going. Follow us, leave a like, or write a positive comment. We read every one, and they are what keeps us going.
Read next: Gemini 3.7 Flash Halves Its Price Until January | Sonnet 5.5 Keeps Its Price And Cuts The Bill
