Gemini 4 Argon: Features, Price and Availability Explained

Updated: 1 October 2026

Google announced Gemini 4 Argon on 30 September 2026. It is the company’s new frontier AI model from Google DeepMind, built for long, multi-step work in software engineering, enterprise tasks such as legal and finance, and cybersecurity.

It is also the first major Gemini release in close to a year, after Google dropped a planned Gemini 3.5 Pro launch. The catch: almost nobody can use it yet. For now, Argon is limited to a group of vetted cyber defenders in Google’s Fairwind Program and to Google’s own teams. Paid API customers and Google AI Ultra subscribers are next in line, but Google has not given a date.

Gemini 4 Argon: Quick Facts

Detail	Information
Model	Gemini 4 Argon
Company	Google DeepMind (Google)
Announced	30 September 2026
Model family	Gemini 4
Key focus	Software engineering, enterprise knowledge work (legal, finance), cybersecurity defence
Output limit	Up to 1 million tokens per response (up from 64,000)
Input context window	Not stated in Google’s announcement (reported as 1 million tokens by The Decoder)
Current availability	Trusted cyber defenders in the Fairwind Program and internal Google teams
API availability	Not yet; paid API customers are first in the planned rollout
Consumer availability	Not yet; Google AI Ultra subscribers are first in the planned rollout
Pricing (API)	$2 / 1M input tokens and $10 / 1M output tokens (introductory); $4 and $20 after that

What Is Gemini 4 Argon?

Gemini 4 Argon is a large AI model — the kind of system that powers chatbots like Gemini and ChatGPT — but it is aimed at heavier work than answering questions. Google pitches it as a model that can stay on a task for a long time: reading through a large codebase, planning a fix, writing it, and checking it, without a person guiding every step.

It sits at the top of Google’s Gemini line-up and follows the Gemini 3 generation, which launched in November 2025. Google has not announced any other Gemini 4 variants yet, such as smaller “Flash” or “Lite” versions.

In Google’s announcement, Koray Kavukcuoglu, Chief AI Architect and SVP of Google DeepMind, describes it as “built to sustain deep reasoning across complex, long-horizon workflows.” Right now, the intended users are security professionals and Google’s own engineers. Regular users will get it later.

Gemini 4 Argon Features

Everything in this section is based on what Google has said and shown. Independent testing is still limited, so treat the performance claims as Google’s claims.

Advanced reasoning and long-horizon tasks

“Long-horizon” simply means tasks that take many steps and a long time, where the model has to remember what it did earlier and plan what comes next. Think of a code migration across hundreds of files rather than a single question and answer.

A related change is output length. Argon can generate up to 1 million tokens in one response, up from 64,000 in earlier Gemini models. A token is roughly a word or part of a word. That bigger output budget matters for jobs like rewriting large chunks of code in one go.

Coding and software engineering

Google shared examples from its own use of Argon. According to the announcement, Google is using it to help migrate C and C++ code to Rust, including parts of the Zircon kernel in its Fuchsia operating system. In one project on the libgav1 video decoder, Argon agents replaced around 32,000 lines of low-level code, and the result ran 2.7 times faster than an earlier Rust port, VentureBeat reported.

On benchmarks, Google reports 77.9% on DeepSWE v1.1, a software engineering test. The picture is not one-sided, though. Reuters reported that Argon trailed rivals on two of four coding-related benchmarks Google published, and Axios reported that internal employee testing showed mixed results.

Cybersecurity

This is where Argon differs most from earlier Gemini models. Google says the model can find software vulnerabilities, confirm they are real, and write patches for them, with much of that done on its own.

Google’s example: through security company Wiz’s Scan for Good programme, Argon “uncovered a critical vulnerability exposing sensitive personal information across healthcare software used by hospitals worldwide,” a flaw Google says previous frontier models had missed. The affected software was not named. On CWE-bench v1, a vulnerability benchmark, Google reports Argon at 68%, tied for first place.

Enterprise knowledge work

Google specifically names legal and finance work, along with research and document-heavy tasks. The company says Argon is strong where work involves visual information, such as analysing professional charts, picking out details from long videos and acting on a series of documents. Google reports a score of 91.7% on LVBench, a long-video understanding test.

Gemini 4 Argon Context Window: What Has Been Announced

Google’s announcement does not state Argon’s input context window. What it does confirm is the output limit: up to 1 million tokens per response. The Decoder reports that the input window is 1 million tokens, unchanged from the Gemini 3 generation.

The context window is how much material a model can read and keep in view at once. A 1-million-token window is roughly several thousand pages of text. In practice, a large window helps with:

Long documents: reading an entire contract set or annual report instead of pasting it in pieces.
Large codebases: looking at many files together, so a fix in one place does not break something elsewhere.
Comparing files: checking several versions of a policy or several research papers side by side.
Long research sessions: keeping earlier findings in view across a long task.

A big window does not guarantee good answers. Models can still miss details buried deep in long inputs, and longer inputs cost more to process.

Gemini 4 vs Gemini 3: What’s Different

Google has published few like-for-like numbers between Argon and Gemini 3 models. This table sticks to what has been documented.

Feature	Gemini 4 Argon	Gemini 3 generation
Reasoning	Pitched for long-horizon, multi-step workflows. Artificial Analysis ranks it 23 points above Gemini 3.1 Pro Preview on its Intelligence Index (via The Decoder)	Not directly comparable from the information currently published
Coding	Google reports 77.9% on DeepSWE v1.1	Not directly comparable from the information currently published
Output limit	Up to 1M tokens	64K tokens
Input context	Not stated by Google; reported as 1M	1M tokens (Gemini 3 Pro)
Multimodal	Input: text, images, video, audio; text output (The Decoder)	Multimodal input (text, images, video, audio)
Agentic workflows	Google reports 51.3% on AutomationBench, first place	Not directly comparable from the information currently published
Cybersecurity	Google says it outperformed Gemini 3.8 Flash Cyber on Wiz’s penetration-testing benchmark	Gemini 3.8 Flash Cyber is the current Fairwind model
Availability	Fairwind Program and internal Google teams only	Generally available in the Gemini app and API

Gemini 4 Argon vs GPT-6 Astra and Claude Opus 5.5

Google compared Argon with OpenAI’s GPT-6 Astra and Anthropic’s Claude Opus 5.5. The numbers below are Google-reported, as summarised by VentureBeat. They have not been independently verified, and OpenAI and Anthropic may report different results for their own models.

Benchmark (Google-reported)	Gemini 4 Argon	GPT-6 Astra	Claude Opus 5.5
DeepSWE v1.1 (coding)	77.9%	74.1%	74.2%
AutomationBench (agentic)	51.3%	41.4%	42.5%
CWE-bench v1 (security)	68%	68%	67%
LVBench (long video)	91.7%	87.5%	83.7%
Terminal-Bench 4.0 (coding)	57.4%	—	66.4%
FrontierSWE v2 (coding)	55.0%	65.5%	—

According to VentureBeat, Argon led outright on 12 of the 18 benchmarks Google disclosed. GPT-6 Astra led on three and Claude Opus 5.5 on two. On the independent Artificial Analysis Intelligence Index, The Decoder reported Argon at 53 points, level with GPT-6 Astra and below Claude Opus 5.5 at 58.

API pricing (per 1M input / output tokens): Gemini 4 Argon $2 / $10 introductory, then $4 / $20; Claude Opus 5.5 $4 / $20; GPT-6 Astra $10 / $50 (VentureBeat). One caveat: The Decoder found Argon used more than twice as many output tokens per task as GPT-6 Astra in its tests, so cheaper per-token rates do not automatically mean a cheaper bill.

Is Gemini 4 Argon Available? Who Can Use It Now

Available now

Trusted cyber defenders in Google’s Fairwind Program, who get the model without its cyber guardrails for defensive work.
Google’s internal teams, who Google says are already using it for debugging and codebase migrations.
Participants in the US government’s voluntary pre-release model access process.

Expected / planned

Google says it will make Argon available to developers, enterprises and consumers “as soon as possible”, starting with paid API customers and Google AI Ultra subscribers.

Not announced

A release date for the API, the Gemini app or Google AI Ultra.
Whether Google AI Pro or free Gemini users will get access.
Free access of any kind.
India-specific pricing or availability.

So, if you are wondering how to use Gemini 4 today: unless your organisation is part of Fairwind, you can’t yet.

Gemini 4 Argon Price

Google has announced API pricing. Argon will launch at an introductory price of $2 per million input tokens and $10 per million output tokens, with cached input tokens charged at 95% off the input price. After the introductory period, pricing rises to $4 per million input tokens and $20 per million output tokens.

Google has not said how long the introductory period will last. It has also not announced how Argon will be priced inside consumer plans, beyond saying Google AI Ultra subscribers will be among the first to get it.

Gemini 4 Argon and Cybersecurity: Why Google Is Going Slow

Google launched the Fairwind Program on 2 September 2026 as a limited-access programme for governments and trusted partners to use its cyber defence tools. Before Argon, the programme was built around Gemini 3.8 Flash Cyber and CodeMender, Google’s AI patching agent.

Google says Fairwind has more than 650 partners worldwide. It prioritises government agencies, critical infrastructure operators in areas like healthcare, telecom, energy and finance, and large technology platforms. Participants must limit access to security, incident response or penetration testing teams, use multi-factor authentication and work in secure cloud environments.

The reason for the slow rollout is the dual-use problem. A model that can find and verify serious vulnerabilities could help attackers as much as defenders. “Safely releasing frontier capabilities at this level requires a phased approach,” Kavukcuoglu said, according to Reuters.

For the wider release, Google says it is strengthening four kinds of safeguards: refusing requests that would help cyberattacks or chemical, biological, radiological and nuclear (CBRN) attacks while still supporting legitimate research; resistance to indirect prompt injection, where hidden instructions in a document or web page try to hijack the model; monitoring the model’s reasoning and actions for misalignment and stopping it when needed; and sealed sandbox environments during high-risk training.

Google is not alone here. Reuters noted that Anthropic has similarly restricted its Claude Mythos Preview model to a small set of trusted organisations.

What Gemini 4 Argon Means for Different Users

Most of this applies once Argon reaches the API and Google AI Ultra. Until then, these are realistic uses based on what Google has shown.

Students

Very little changes for now — Argon is not in the free Gemini app. When it does arrive, the long-video and chart analysis Google highlights could help with tasks like reviewing a recorded lecture or reading data-heavy research papers. Expect it to sit behind paid plans at first.

Developers

This is the group with the most to gain. The 1M-token output limit means a single response can carry far more code. Large migrations, refactoring legacy code, or generating full test suites are the kind of jobs Google has demonstrated internally. Watch for API access and the end of introductory pricing.

Businesses

Legal and finance teams are named directly by Google. Practical examples include reviewing large contract sets, checking financial filings across several documents, and automating multi-step back-office workflows. Security teams in critical sectors can ask about Fairwind eligibility now.

Researchers

Google says Argon helped its quantum computing researchers beat a published optimisation baseline by 40%. Researchers working with long datasets, papers or video material should look at the long-context and long-video claims once access opens.

Content creators

Google has not positioned Argon as a creative tool. Its stated strengths are reasoning, code and analysis. Creators may find it useful for research and analysing long video, but there is no announced image or video generation in Argon.

What Is Still Unknown About Gemini 4 Argon

Release dates for the API, Google AI Ultra and the Gemini app.
The input context window, which Google’s announcement does not state.
How long introductory pricing lasts, and consumer plan pricing.
Access for Google AI Pro and free users.
Other Gemini 4 models, such as Flash or Lite variants.
Broad independent benchmarks. Most published scores come from Google.
Product integrations like Search, Workspace or Android.

Frequently Asked Questions

What is Gemini 4 Argon?

Gemini 4 Argon is Google DeepMind’s newest frontier AI model, announced on 30 September 2026. Google built it for long, multi-step tasks in software engineering, legal and finance work, and cybersecurity. It is the first model in the Gemini 4 family.

When was Gemini 4 Argon released?

Google announced Gemini 4 Argon on 30 September 2026. It was released the same day only to trusted cyber defenders in Google’s Fairwind Program and to Google’s internal teams. A wider public release date has not been announced.

Is Gemini 4 available to the public?

No. As of 1 October 2026, Gemini 4 Argon is limited to vetted cybersecurity partners in the Fairwind Program and Google’s own teams. Google says paid API customers and Google AI Ultra subscribers will get access first, but has not given a date.

How can I use Gemini 4?

Most people can’t use it yet. Access currently runs through Google’s Fairwind Program, which is open to governments, critical infrastructure operators and selected security partners. Once it rolls out, the first routes will be the paid Gemini API and a Google AI Ultra subscription.

Is Gemini 4 free?

No free access has been announced. Google has published API pricing of $2 per million input tokens and $10 per million output tokens during an introductory period, rising to $4 and $20 later. Consumer access is expected to start with the paid Google AI Ultra plan.

What is Gemini 4’s context window?

Google has not stated Argon’s input context window in its announcement. It has confirmed an output limit of up to 1 million tokens per response, up from 64,000. The Decoder reports the input window is 1 million tokens, the same as Gemini 3.

What can Gemini 4 Argon do?

According to Google, Argon can handle long coding projects, analyse charts and long videos, work across many documents, and find, verify and patch software vulnerabilities. Google has shown it migrating C and C++ code to Rust and finding a critical flaw in hospital software.

Is Gemini 4 better than Gemini 3?

Google says yes on its own tests, and Artificial Analysis ranks Argon well above Gemini 3.1 Pro Preview. The clearest documented change is output length: 1 million tokens versus 64,000. Many other direct comparisons have not been published yet.

Is Gemini 4 available through the API?

Not yet. API pricing has been announced, and Google says paid API customers will be among the first to get access, but there is no API release date. For now, only Fairwind Program participants and internal Google teams can use it.

Does Gemini 4 support coding?

Yes. Software engineering is one of Argon’s three main focus areas. Google reports 77.9% on the DeepSWE v1.1 benchmark, though Reuters noted it trailed rivals on two of four coding benchmarks Google published.

Why is Google limiting access to Gemini 4 Argon?

Because of its cybersecurity abilities. A model that can find and confirm serious software flaws could be misused by attackers. Google says it is gathering feedback from defenders and strengthening safeguards before releasing Argon more widely.