Google has released Gemini 4 Argon, its new frontier model, to vetted cybersecurity teams only. More than 650 partners in Google’s Fairwind Program got access on 30 September 2026; paid API customers and Google AI Ultra subscribers come next. Google has not given a date for them.
Who Can Use Gemini 4 Argon Today
Access runs through the Fairwind Program, a restricted-access scheme Google opened on 2 September 2026, and nothing else. Koray Kavukcuoglu, senior vice-president of Google DeepMind and the company’s chief AI architect, announced Argon on Google’s blog on 30 September 2026 and said it is rolling out to a set of trusted cyber defenders through the Fairwind Program.
Fairwind is not open to applications from the general public. Four Flynn, Google’s vice-president of security and privacy, set out the eligible organisations when the programme launched:
- Governments and national cyber authorities: state bodies responsible for defending public networks.
- Critical infrastructure operators: sectors such as healthcare, telecommunications, energy and finance.
- Core technology platforms: large providers whose software other organisations depend on.
- Google Cloud customers and security partners: firms Google has separately vetted.
Membership alone does not put Argon in front of an employee. Google’s Fairwind Program announcement states that “we have more than 650 participating partners globally”, and that participants must accept “strict operational standards, including limiting access to employees within their internal cybersecurity, incident response, or penetration testing teams and deploying protections like multi-factor authentication.” An engineer or analyst at a member organisation who does not sit in one of those three teams is not covered.
Gemini 4 Argon Pricing, and the Date Google Did Not Give
Google has published the prices but not the schedule. The rates below come from Google’s Gemini 4 Argon announcement, which does not state when the introductory period ends.
| Item | Introductory rate | Standard rate |
|---|---|---|
| Input tokens | $2 per million | $4 per million |
| Output tokens | $10 per million | $20 per million |
| Cached input tokens | 95% off the input price, or $0.10 per million | 95% off the input price, or $0.20 per million |
Google’s stated plan is that Argon becomes available to developers, enterprises and consumers as soon as possible, with paid API customers and Google AI Ultra subscribers first in line. No calendar date accompanies that commitment. The round of AI API price cuts in September 2026 is a reminder of how quickly these rates move once a model reaches general availability.
As of 2 October 2026, Gemini 4 Argon is available only to Fairwind Program partners. Google has set no date for paid API customers, Google AI Ultra subscribers or consumers, and has not said when the introductory pricing expires.
Why Trusted Defenders Get Argon Without Cyber Guardrails
Google is giving the restricted group its most capable cyber model with the safety limits on cyber tasks removed, which inverts the usual order of a staged release. In Google’s words, Argon goes to trusted defenders “without cyber guardrails so they can leverage its full frontier-level cybersecurity defense capabilities.” The narrow audience gets the less restricted version while the wide audience waits.
The reasoning is that the same capability cuts both ways. A model good enough to find and patch a software flaw on its own is also a model good enough to find one and leave it open, so Google is handing it first to the people whose job is the patching. Google says the deployment carries “misalignment mitigations that monitor Argon’s chain-of-thought and actions and stop execution when necessary” — monitoring of the model’s internal reasoning trace, with the ability to halt a task mid-run. The same tension surfaced when Gemini breached three companies during a controlled security test earlier in 2026.
What Google Reported on the Benchmarks
Every score below is Google’s own, published in its announcement, and none has been independently reproduced. Google positions Argon on software engineering, enterprise knowledge work such as legal and finance, and cybersecurity defence.
| Benchmark | What it measures | Argon score |
|---|---|---|
| DeepSWE v1.1 | Real-world software engineering tasks | 77.9% |
| Vals Index | Enterprise knowledge work | 68.9% |
| AutomationBench | Multi-step office automation | 51.3%, ranked first |
| CWE-bench v1 | Repairing known classes of software vulnerability | 68%, tied first |
| LVBench | Long video understanding | 91.7% |
Google also published side-by-side comparisons against OpenAI’s GPT-6 Astra and Anthropic’s Claude Opus 5.5 across those categories. Separately, The Hacker News reported that Google’s evaluation placed Argon first on Gray Swan’s indirect prompt injection benchmark, a test of whether a model can be hijacked by instructions hidden inside the content it reads. That ranking has been reported by a single outlet so far.
The headline specification is the output limit, which rises to one million tokens in a single response from 64,000 in earlier Gemini models. That is the change that makes the long-running, multi-step work Google describes practical, because the model no longer has to stop and be restarted partway through a large job. Earlier releases in the line, such as Gemini 3.8 Live and its published voice benchmarks, were priced and measured on far smaller workloads.
What Changes for Gemini App and AI Ultra Users
Nothing changes today. Gemini 4 Argon does not appear in the Gemini app, and a Google AI Ultra subscription does not unlock it, even though Ultra subscribers are named as part of the next wave. Developers cannot call it either, because the rollout has not reached paid API customers.
Fairwind partners were already running Gemini 3.8 Flash Cyber and CodeMender, Google’s automated patching harness, before Argon arrived. For everyone outside that group, the practical result of the 30 September announcement is a published price list for a model they cannot yet buy.
What Is Still Unconfirmed
- The wider release date: Google says as soon as possible and names no day, month or quarter.
- The end of introductory pricing: the $2 and $10 rates and the $4 and $20 rates that replace them are both published, but not the date of the switch.
- Independent benchmark results: the scores in circulation are Google’s own, and no third-party evaluation of Argon has been published.
- Consumer availability: Google names developers, enterprises and consumers as eventual audiences without saying in what order consumers arrive, or through which products.
Frequently Asked Questions
Can I Use Gemini 4 Argon Today?
Only if your organisation is a Fairwind Program partner and you work in its internal cybersecurity, incident response or penetration testing team. Google has not opened Argon to the Gemini app, to developers or to consumers.
How Much Does Gemini 4 Argon Cost?
Google’s introductory rates are $2 per million input tokens and $10 per million output tokens, rising to $4 and $20 after the introductory period. Cached input tokens are discounted by 95 percent.
When Will Gemini 4 Argon Reach Google AI Ultra Subscribers?
Google has not given a date. It says paid API customers and Google AI Ultra subscribers are first in line once the rollout widens beyond Fairwind partners.
What Is the Fairwind Program?
Fairwind is a restricted-access Google programme opened on 2 September 2026 for governments, critical infrastructure operators and vetted security partners. Google says it has more than 650 participating partners globally.
How Many Tokens Can Gemini 4 Argon Produce?
Up to one million tokens in a single response, up from 64,000 tokens in earlier Gemini models. Google describes this as an industry-leading output limit.
