Is Gemini 4 Argon a Pro model?

No. Google calls Argon its new frontier model, not a Pro model. The accurate statement is that Secret Chat AI's flagship Gemini slot has a frontier model behind it again.

Which Secret Chat AI modes use Gemini 4 Argon?

Research runs on Gemini 4 Argon, with Google Search grounding and URL context. Smart Agentic also escalates questions that need deep thinking or research to the Argon flagship tier automatically.

Does Web Search use Gemini 4 Argon?

No. Web Search remains on Gemini 3.8 Flash with Google Search grounding and URL context. Fast Chat, routing, chat titles and summaries remain on Gemini 3.5 Flash-Lite.

Do the council panels use Argon?

No. The Gemini seat stays on Gemini 3.5 Flash-Lite for a light council panel and Gemini 3.8 Flash for a mid panel.

Do the Work generators use Argon?

Not by default. The Gemini Work generators write on Gemini 3.8 Flash, but on the agentic generators — all except Email, Translator and Rewriter — a turn set to deep research or one the Smart router judges to need deep thinking reaches the flagship tier, now Gemini 4 Argon.

What can Gemini 4 Argon accept as input?

Google's model page lists text, images, video, audio and PDF input, with text output. It has a unified 1,048,576-token input-and-output context limit, with 65,536 output tokens by default and up to 1,048,576 through Google's long-decode option.

Is the unrestricted cyber version of Argon available on Secret Chat AI?

No. The standard model blocks offensive tasks such as pentesting and exploitation. Google limits the version without those cyber guardrails to vetted defenders in its Fairwind Program, and that version is not available on Secret Chat AI.

Do I need a Google account to use Gemini 4 Argon?

No. Secret Chat AI sends queries without a Google account or profile, stores no server-side chat archive and never uses chats for training. Prompts and attachments still go to Google's API verbatim, provider retention may apply, and you are responsible for removing identifying details before sending.

Is Gemini 4 Argon faster than Gemini 3.8 Flash?

We have not measured its speed or latency, so we do not claim that it is faster or slower. A frontier model doing deep, multi-step reasoning is expected to take longer than Flash on demanding work, but that expectation is not a measurement.

Gemini 4 Argon on Secret Chat AI

Published October 14, 2026

Gemini 4 Argon is now live on Secret Chat AI. It powers Gemini's Research mode and the hardest questions that Smart Agentic escalates on its own. Google's newest frontier model is finally behind the flagship slot again.

That last word matters: flagship, not Pro. Google does not call Argon a Pro model, and neither will we. It is a new name and a new generation, arriving after an odd stretch in which Google's fast models kept moving while its heaviest general-purpose tier did not.

This is the third chapter in a story that began with Gemini 3.7 Flash and the Pro tier that never came, then continued when Gemini 3.8 Flash shipped in September. Since August 14, Secret Chat AI has put the newest Flash model in the flagship slot rather than leave hard questions on gemini-3.1-pro-preview, a January build still labelled preview and never given a successor. As of today, that stopgap is over.

What Is Gemini 4 Argon?

Google announced Gemini 4 Argon on September 30. Koray Kavukcuoglu of Google DeepMind called it Google's “new frontier model,” built for real-world software engineering, enterprise knowledge work such as legal and finance, and cybersecurity defense. More tellingly, Google says it was designed “to sustain deep reasoning across complex, long-horizon workflows.” The first rollout was deliberately narrow, through Google's Fairwind Program for trusted cyber defenders, before developer access followed. That history and wording come from Google's announcement.

Developers gained access on October 13 through the paid tier of the Gemini API in Google AI Studio and Android Studio, according to Google's developer announcement. Google also lists Argon in Gemini for business, Google AI Studio, Google Antigravity and its cloud API under the model ID gemini-4-argon in the Google Cloud release. That is the stable API model now running in the flagship Gemini slot on Secret Chat AI.

The official model page gives the useful technical shape:

A unified 1,048,576-token context window. Input and output share that limit. The normal synchronous output ceiling is 65,536 tokens; Google's long-decode option can extend output as far as the full 1,048,576-token limit.
Text out, many formats in. Argon accepts text, images, video, audio and PDFs, and produces text.
Three thinking levels. The API supports low, medium and high. Secret Chat AI sends no thinking level, so Google's default applies.
Tools for research and structured work. Google lists Search grounding, URL context, code execution, function calling and structured outputs among the supported capabilities.

The million-token figure needs one small correction to the usual shorthand. It is not simply “a million tokens in and another million out.” It is a shared input-and-output budget. Even so, it is enough room for the kind of assignment Argon is meant to sustain: several documents, several dependent steps and a substantial finished answer rather than a quick reply to one isolated question.

Built for Work That Has a Middle

Most model announcements promise harder reasoning. Argon's more specific promise is persistence: not merely getting the first step right, but carrying assumptions, evidence and intermediate results through a long chain without losing the plot. That is why Google's examples include using Argon agents to turn an existing Rust port of libgav1, its open-source video decoder, into a memory-safe decoder that runs 2.7× faster than that port, finance-agent work and legal research and drafting rather than ordinary chat.

Google's benchmark claims follow the same pattern. It reports 77.9% on DeepSWE v1.1, which it calls a new state of the art for real-world long-horizon software engineering; 51.3% and first place on Zapier's AutomationBench; and a leading result on the Vals Index, which Google describes in its Cloud release as measuring work across finance, coding, legal and tax. Those results appear in Google's own announcement. We did not re-run those evaluations, so they are Google's claims, not ours.

There is another result worth noticing because Research can open pages you link. Google says Argon leads Gray Swan's Indirect Prompt Injection benchmark and reduced attack success across 15 adaptive attempts to 0.7%. An indirect prompt injection is a hostile instruction hidden in an email, document or web page the model reads. Google calls Argon its most resilient model yet against that class of attack in the Google Cloud post. That is a benchmark result, not a guarantee that hostile content cannot fool it.

We have not measured Argon's speed or latency, and we will not invent a number. A frontier model reasoning at depth should be expected to take longer than Flash on demanding work. That is an expectation about the job it is doing, not a measurement of this deployment. If the question needs one current fact, Web Search remains the more natural mode. If it needs a long argument assembled and checked across sources, that is what the flagship slot is for.

What Changed on Secret Chat AI

Gemini still appears as one entry in the model picker. The round mode button beside the message box offers Smart Agentic, Fast Chat, Web Search, Research and Create Image. You do not need to hunt for a second Argon entry.

Two paths now reach Gemini 4 Argon:

Research runs on gemini-4-argon, with Google Search grounding and URL context attached. It can search for current sources and open pages you link.
Smart Agentic, the default, escalates questions that need deep thinking or research to that same flagship tier. Its router runs on Gemini 3.5 Flash-Lite, reads the message, and chooses the handling.

Smart Agentic is therefore not a model name. It is traffic control. Quick answers stay on Gemini 3.5 Flash-Lite. A current lookup goes to Gemini 3.8 Flash with Web Search. A question that needs the deepest route now goes to Argon. You can let the router choose or select Research yourself when you already know the assignment is substantial.

Gemini is a paid model family on Secret Chat AI, used from a pay-as-you-go US-dollar balance. Gemini is not part of the free daily quota.

What Did Not Change

A flagship upgrade is easy to overstate. Argon has a precise job here, and the rest of the Gemini family stays where it was:

Web Search remains on Gemini 3.8 Flash, with Google Search grounding and URL context.
Fast Chat, the Smart Agentic router, chat titles and summaries remain on Gemini 3.5 Flash-Lite.
The AI Council's Gemini seat remains on the Flash tiers: Gemini 3.5 Flash-Lite on a light panel and Gemini 3.8 Flash on a mid panel. Councils seat each family's mid tier, not its heaviest model. A council is the alternative to betting on one lab's roadmap: it sends the question to every model on the panel, and agreement between them is evidence, not proof.
Work generators write on Gemini 3.8 Flash by default. On the agentic generators — all except Email, Translator and Rewriter — a turn set to deep research or one the Smart router judges to need deep thinking reaches the flagship tier, now Gemini 4 Argon.
Create Image remains on Nano Banana, a separate image model.

The cyber version needs the same boundary. The public Argon API has mitigations that block offensive work such as pentesting and exploitation. Google makes an unrestricted version available only to vetted defenders through Fairwind, as its cloud announcement explains. That unrestricted variant is not on Secret Chat AI, just as Gemini 3.8 Flash Cyber was not here last month.

How to Use Gemini 4 Argon

Select Gemini in the model picker, then either leave Smart Agentic on or choose Research with the round button beside the message box. Research is the deliberate choice when you have a long source set, linked material or a conclusion that depends on several earlier steps being right.

Argon is wasted on “summarise this in five bullets.” Give it the awkward middle of the assignment: reconcile the numbers, trace each conclusion to evidence, expose conflicts, and say what cannot be established. Here is a copy-ready example:

I am going to provide a set of public filings, published reports and linked sources about a company. First build a dated fact table and identify conflicts or missing evidence. Then test each material claim about the company's performance and obligations against those public sources, showing the source for every finding. Next write a risk analysis that separates verified facts, assumptions and unresolved questions. Finally produce a decision memo with the three issues that could change the recommendation, the additional public source or fact needed for each one, and a step-by-step review checklist. Do not treat silence as evidence, do not invent missing terms, and pause to tell me if a later step cannot be completed because an earlier dependency is unresolved.

This is analysis assistance, not a substitute for a qualified lawyer, accountant or other professional. More practically, it is also a task where the documents themselves may identify people. Read the privacy limits below before sending them.

Private by Design

The model changed. The privacy arrangement did not. Secret Chat AI stores no chat archive on its side: history lives in your own browser, and a prompt exists on our servers only as long as it takes to fetch the answer. There is no Google account to connect, no profile built around your use and no chat tied to you as a person. Queries reach Google as a stranger's, and they are never used for training.

But anonymity is about the account link, not the contents. Your prompts and attachments go to Google's API verbatim. Provider retention may still apply. If a contract names a client, a spreadsheet contains an account number or a PDF includes a signature, those details remain in what Google receives. Redacting identifying details before sending is your responsibility.

That means “private” is not a promise that data cannot leak, and it does not make sensitive material automatically suitable for upload. It means Secret Chat AI does not keep the conversation or attach it to your identity. The fuller explanations are in Does Gemini Read Your Data?, Private Gemini and our account of local browser chat storage.

Conclusion

For two months, the honest answer to “what handles the hardest Gemini questions?” was: the newest Flash model, because the alternative was a stale preview. Today that answer changes. Research and the questions Smart Agentic escalates to the flagship tier now run on Gemini 4 Argon, Google's new frontier model.

The boundary is just as important as the upgrade. Web Search and the council's Gemini seat remain on Flash. The Gemini Work generators write on Gemini 3.8 Flash by default, while a turn set to deep research or one the Smart router judges to need deep thinking reaches Argon on the agentic generators — all except Email, Translator and Rewriter. Image creation remains on Nano Banana. Argon is here for the long job with a middle: the document set, the linked sources and the conclusion that depends on five earlier conclusions holding together.

You can use it now by choosing Gemini on Secret Chat AI and selecting Research, or by leaving Smart Agentic to escalate the hard questions itself.

Frequently Asked Questions

What is Gemini 4 Argon?
Gemini 4 Argon is Google's new frontier reasoning model for complex, long-horizon work across software engineering, legal and finance tasks, and cybersecurity defense. Google announced it on September 30, 2026, and made it available to developers on October 13.
Is Gemini 4 Argon a Pro model?
No. Google calls Argon its new frontier model, not a Pro model. The accurate statement is that Secret Chat AI's flagship Gemini slot has a frontier model behind it again.
Which Secret Chat AI modes use Gemini 4 Argon?
Research runs on Gemini 4 Argon, with Google Search grounding and URL context. Smart Agentic also escalates questions that need deep thinking or research to the Argon flagship tier automatically.
Does Web Search use Gemini 4 Argon?
No. Web Search remains on Gemini 3.8 Flash with Google Search grounding and URL context. Fast Chat, routing, chat titles and summaries remain on Gemini 3.5 Flash-Lite.
Do the council panels use Argon?
No. The Gemini seat stays on Gemini 3.5 Flash-Lite for a light council panel and Gemini 3.8 Flash for a mid panel.
Do the Work generators use Argon?
Not by default. The Gemini Work generators write on Gemini 3.8 Flash, but on the agentic generators — all except Email, Translator and Rewriter — a turn set to deep research or one the Smart router judges to need deep thinking reaches the flagship tier, now Gemini 4 Argon.
What can Gemini 4 Argon accept as input?
Google's model page lists text, images, video, audio and PDF input, with text output. It has a unified 1,048,576-token input-and-output context limit, with 65,536 output tokens by default and up to 1,048,576 through Google's long-decode option.
Is the unrestricted cyber version of Argon available on Secret Chat AI?
No. The standard model blocks offensive tasks such as pentesting and exploitation. Google limits the version without those cyber guardrails to vetted defenders in its Fairwind Program, and that version is not available on Secret Chat AI.
Do I need a Google account to use Gemini 4 Argon?
No. Secret Chat AI sends queries without a Google account or profile, stores no server-side chat archive and never uses chats for training. Prompts and attachments still go to Google's API verbatim, provider retention may apply, and you are responsible for removing identifying details before sending.
Is Gemini 4 Argon faster than Gemini 3.8 Flash?
We have not measured its speed or latency, so we do not claim that it is faster or slower. A frontier model doing deep, multi-step reasoning is expected to take longer than Flash on demanding work, but that expectation is not a measurement.