Gemini 4 Argon Starts With Cyber Defenders

Google has announced a new frontier Gemini model, but most UK teams cannot pick it in the API yet.

What happened

On 30/09/2026 Google DeepMind introduced Gemini 4 Argon.

The company positions Argon as a frontier model for complex, long horizon workflows across software engineering, enterprise knowledge work such as legal and finance, and cybersecurity defence.

Day one access is not a public model picker flip.

Argon is rolling out first to a set of trusted cyber defenders through Google's Fairwind Program.

Google says it is taking a phased approach, including engagement with the U.S. government's voluntary process for pre release model access, while it gathers feedback and hardens guardrails.

Broader availability is promised for developers, enterprises, and consumers as soon as possible, starting with paid API customers and Google AI Ultra subscribers.

No public calendar date or public model id appears in the announcement.

The headline engineering change is output length.

Argon's output token limit rises to 1 million tokens, up from the previous 64,000 token ceiling Google cites for earlier models.

That gives the model room to think and write far longer trajectories in a single run.

Introductory list pricing is 2.00 US dollars per million input tokens and 10.00 US dollars per million output tokens.

Cached input tokens are priced at a 95 percent discount to the input rate.

At roughly 0.75 pounds to the dollar that is about £1.50 input, £7.50 output, and about £0.08 for cached input per million tokens.

After the introductory period those rates rise to 4.00 US dollars and 20.00 US dollars per million tokens for input and output respectively, or about £3.00 and £15.00.

Google reports Argon at 77.9 percent on DeepSWE v1.1, a long horizon software engineering bench.

On Zapier's AutomationBench it ranks first at 51.3 percent.

On LVBench for long video understanding it scores 91.7 percent.

On CWE-bench v1 for remediating security vulnerabilities it ties for first at 68 percent.

Those figures are vendor reported, so treat them as a signal to verify later on your own workloads.

Inside Google, Argon is already used for specialised coding, research, and writing.

Examples in the launch post include quantum algorithm optimisation that beat a published baseline by 40 percent in minutes, fleet wide memory optimisation that freed more than 300 TiB once rolled out, and large scale C and C++ to Rust migrations with heavy audit before production.

Wiz is named as an early Fairwind user applying Argon to vulnerability discovery on critical infrastructure software.

Why this matters for UK businesses

Plenty of UK product, agency, and security teams already budget for Gemini API calls or Google AI Ultra seats.

A frontier model that is not yet on your bill is still a planning event.

Finance needs the introductory and post introductory price bands before someone assumes Argon will land at Flash class rates.

Security and compliance leads need to know that the first external cohort is cyber defence oriented, with a version for trusted defenders that Google describes as running without the cyber guardrails applied to broader releases.

That is useful context if you are a critical infrastructure operator considering Fairwind, and a reminder not to expect the same unconstrained cyber tooling on day one of a consumer or standard API rollout.

The 1 million token output window matters for migration work, long research briefs, and agent loops that previously hit short generation caps.

It does not magically remove context costs on the input side, and a long thinking run can still burn a lot of output tokens at 10.00 US dollars per million.

How we see it at Adevious AI

We would not rewrite production routes today.

There is no public model id to swap in, and Google is explicit that safeguards are still being strengthened before broad availability.

What we would do is mark Argon as the next Gemini frontier tier in the roadmap, with two gates: published API access for paid customers, and a measured pilot on non critical coding and knowledge work tasks.

Watch the introductory pricing window closely.

A jump from 2.00 and 10.00 US dollars to 4.00 and 20.00 US dollars per million tokens doubles the unit cost after the promo period, which can wreck a fixed quarterly forecast if someone loads Argon into every agent path.

Also separate the cyber defence story from the general coding story.

Fairwind access and Scan for Good style use cases are not the same product surface as Gemini in Sheets or a standard chat subscription.

If your firm is not an eligible defender, treat the cyber benchmarks as background, not a feature you can enable this week.

What to do this week

If you already use the Gemini API, keep your current production model pinned and open a short watch list for Argon's model id in Google AI Studio and the official model docs.

If you are on Google AI Ultra, check the model picker over the coming days, but do not assume Ultra access has shipped until Google says so in product.

If you operate critical systems and may qualify for Fairwind, review Google's Fairwind Program page and apply through the official form rather than waiting for a consumer launch blog.

Price a typical million token block in GBP at both the introductory and post introductory bands before anyone prototypes an Argon heavy agent.

Bottom line

Gemini 4 Argon is a real frontier announcement with a clear price card and a serious long output window.

It is not yet a drop in upgrade for most UK Gemini users.

Plan for paid API and Ultra access next, keep an eye on the Fairwind path if you are a defender, and do not change production defaults until Google publishes a model id you can call.

Sources

Google DeepMind / Google, Introducing Gemini 4 Argon, 30/09/2026: https://blog.google/innovation-and-ai/models-and-research/gemini-models/gemini-4-argon/

Google DeepMind, Fairwind Program: https://deepmind.google/fairwind-program/

Acronyms

API: Application Programming Interface

CWE: Common Weakness Enumeration

GBP: Great British Pound

SME: Small and Medium sized Enterprise

TiB: Tebibyte

UK: United Kingdom

US: United States