Gemini 4 Argon: TechCrunch calls it Google's most powerful model, and you cannot use it

Google announced Gemini 4 Argon on September 30, 2026, in a post on its own blog by Koray Kavukcuoglu, SVP of Google DeepMind and Chief AI Architect at Google. TechCrunch's Lucas Ropek reported the release the same day and described it as the company's most powerful model yet. For anyone running a business, the point is simple. You will hear a lot about this model and you cannot switch it on today, so the useful question is what it signals about the tools you will be offered next.

Argon is real, but almost nobody can use it yet

According to Google's post, Argon is rolling out to "a set of trusted cyber defenders" through what Google calls its Fairwind Program. TechCrunch describes Fairwind as Google's security initiative and says the model is only being rolled out to a select group of the company's cyber partners. If you run a clinic, a shop or a trades company, you are not in that group, and there is nothing to sign up for today.

Google says the wider release will come in stages. Its post says the company is taking part in the U.S. government's voluntary process for pre-release model access. Google says it will keep gathering feedback from early testers while it works on guardrails, and will then make Argon available to developers, enterprises and consumers "as soon as possible." When that happens, Google says it will start with paid API customers and Google AI Ultra subscribers.

Think of it like a manufacturer that has published the spec sheet and the price of a new commercial oven but has installed it only in its own kitchen and a few partner kitchens. You can read about it and plan around it, but you cannot order one.

The price is published even though the product is not open

Google did publish what Argon will cost developers, which is unusual for a model that is not broadly available. Google's post says Argon "will launch at an introductory price" of $2 per million input tokens and $10 per million output tokens, with cached input tokens priced at 95% off the input token price. Tokens are the units AI companies use to meter what goes into and comes out of a model. This is a price for software builders, not a monthly fee for a shop owner.

The detail most coverage skipped is how much the model can write in one go. Google says it is raising Argon's output token limit to 1 million tokens, up from the previous 64K, and calls that industry-leading. In plain terms, Google is betting that the model can work on one long task for a long time and hand back a very large result, rather than answering in short bursts. That design is aimed at long jobs like reviewing a stack of documents or rewriting a large piece of software, not at quick chat.

Google's benchmark claims point straight at office work

Most of Google's scorecard covers the kind of work a professional practice does. Google says Argon is the leading model on the Vals Index, which Google describes as measuring economic impact across finance, coding, legal and tax work. TechCrunch notes that Google cites Vals, which TechCrunch calls an increasingly popular AI benchmarking startup, to show Argon leading that index. Google also reports leading performance on Vals Finance Agent v2 and on Harvey's Legal Agent Benchmark for legal research and drafting.

The number that maps most closely to running a business comes from AutomationBench, which Google describes as Zapier's benchmark for end-to-end execution across core business functions. Google says Argon ranks first there with a score of 51.3%. Read that both ways: by Google's own account it is the best score on that test, and it is still roughly half. On coding, Google reports 77.9% on DeepSWE v1.1. On long video understanding, it reports 91.7% on LVBench.

TechCrunch adds the competitive frame. In its blog, Google claims Argon scored significantly higher than OpenAI's GPT-6 Astra and Anthropic's Fable and Opus models across a variety of benchmarks. Those comparisons come from Google and were published by Google.

The cybersecurity angle matters most for clinics and anyone holding records

Google says it trained Argon specifically for defensive cybersecurity, and that the model "can autonomously find, validate, and patch critical software vulnerabilities," a line TechCrunch also quotes. Google's post gives one example that should get a clinic owner's attention. Wiz is using Argon through its Scan for Good initiative. According to Google, the model uncovered a critical vulnerability exposing sensitive personal information across healthcare software used by hospitals worldwide, a risk Google says previous frontier models had missed. Google does not name the software in its post.

The practical takeaway is less about Argon itself than about the software you already rent: tools like this are being pointed at the systems that hold your patients' and customers' records. That cuts both ways. Defenders get faster at finding holes, and Google spends a whole section of its post on preventing misuse. Google says the model is designed to refuse harmful requests for cyber attacks. It also says it monitors the model's reasoning and actions and stops execution when the model tries to go beyond what a user intended. Separately, Google says it is releasing Argon without cyber guardrails to trusted defenders and its own internal teams so they can use its full capabilities.

The other side: a lot here is unknown or unpublished. Google has not given a date for wider access beyond "as soon as possible," and its post does not say how long the introductory price lasts. The announcement does not include the price of a Google AI Ultra subscription. Every benchmark figure above comes from Google's own post, and the Wiz example is Google's description, with the software left unnamed. TechCrunch also points out that top AI labs keep racing to release more powerful models even as those same firms warn that AI could spin out of control. Within that race, Google is choosing to hand a version without cyber guardrails to selected defenders.

Your customers will meet this through apps they already use

Most small businesses will never connect to an AI model directly. Their customers will meet it inside apps. Google said on August 11, 2026, that the Gemini app had passed 1 billion monthly users. The same post said small businesses use Gemini's creative features and that the app generates more than 150 million images every day. TechCrunch says Google's user count makes Gemini competitive with OpenAI.

Google's Argon post does not say when, or whether, Argon will power the free Gemini app. It names paid API customers and Google AI Ultra subscribers as first in line. Argon has not reached the general public, so the people asking an AI about your business today are not using it.

The search changes this week will reach you before Argon does

If you want something that touches your business this week, look at search rather than Argon. On September 30, 2026, Search Engine Roundtable's Barry Schwartz reported that Google seems to be showing AI Overviews for many large brand names. That report rests on a practitioner's finding. Chris Long wrote on X that 93% of the brands he tested had AI Overviews appearing, and Schwartz notes the test covered about 100 brands. DemandSphere, also quoted by Schwartz, said AI Overviews on branded queries tripled in September. Schwartz observed that most of these overviews appear lower on the page, below the company's own result, rather than above it.

Schwartz also reported on September 30 that Google Maps now requires people to sign in to a Google account to see all the reviews on a Google Business Profile, to sort them, or to leave one. He says he assumes the change is meant to stop bots, scrapers and AI platforms from collecting reviews, and he presents that as a guess. For a business that leans on its reviews, a visitor who is not signed in may now see only part of what customers have said about you.

Bottom line

With Gemini 4 Argon, Google is claiming leading benchmark results, backed by a published developer price, a very long output limit and a strong pitch for legal, finance, coding and security work. None of it is open to a typical business today. The honest move this week is to watch, not buy: wait for Argon to reach paid API customers and Google AI Ultra subscribers, then see which of the tools you already pay for start using it. In the meantime, you can check two things in search: whether an AI Overview appears when you search your own business name, and what a signed-out visitor sees of your Maps reviews.

When your software vendor tells you their product now runs on the newest model, will you know what to ask them?