Gemini 4 Argon Release Date: Is It Out, and Who Can Use It?

Gemini 4 Argon is available through the Gemini API and Google AI Studio, according to Google's current developer guide, checked October 14, 2026^[1]. Google announced the model on September 30, 2026 with a phased rollout to trusted cyber defenders^[2]. On October 13, Google announced broader global availability for developers, enterprises and eligible consumer subscribers^[3]. The earlier description of developer access as unavailable is now out of date.

For readers asking whether Gemini 4 Argon is out, that distinction matters. You can find a public model ID and developer documentation today. Consumer subscriptions, regional eligibility and the Fairwind security program are separate access questions. This guide separates the announcement history from the developer routes Google documents now, with links to the Gemini 4 Argon model page and request reference on reAPI.

TL;DR

Google announced Gemini 4 Argon on September 30, 2026 with an initial Fairwind rollout^[2].
Google's current guide says the model is available through the Gemini API and Google AI Studio. The model ID is gemini-4-argon^[1].
Google announced broader availability on October 13. The Gemini app and AI Mode in Search are open to Google AI Ultra and Pro subscribers; developer API access uses a separate paid tier^[3].
The 1,048,576-token window is shared by input, thinking and output. Long decode needs continuation beyond individual request limits^[4].
Use the current model documentation to choose your access route; the September announcement alone no longer describes present developer availability.

What happened on September 30

Google introduced Argon as its new frontier model, built "to sustain deep reasoning across complex, long-horizon workflows" in software engineering, enterprise knowledge work such as legal and finance, and cybersecurity defense^[2]. The launch post came with a full benchmark table, a price, and a list of internal results: Argon agents moving C and C++ codebases to Rust, a libgav1 decoder that runs 2.7x faster than the earlier Rust port, and memory optimizations that freed more than 300 TiB across Google's data centers^[2].

The September 30 post did not include a public model ID or developer request reference. Those details are now present in Google's developer documentation. Google described a phased rollout that would use early feedback to improve its guardrails^[2].

So when people search for the Argon release date, they are really asking two separate questions. The announcement date is settled: September 30, 2026. The broader availability announcement followed on October 13, 2026^[3]. October 14 is this article's review date, not a separate release milestone.

Where to access Gemini 4 Argon now

Google names the Gemini API and Google AI Studio as current developer access routes^[1]. Its model page identifies gemini-4-argon and documents text, image, video, audio and PDF input with text output^[5]. These are Google's model capabilities; they do not mean every application built around the model exposes every input type.

Route	What to check before use
Gemini API	Your Google project, billing eligibility and the current model documentation
Google AI Studio	Whether the model is available to your signed-in account and location
reAPI	The model page, current rates and API documentation
Gemini app and AI Mode in Search	Google AI Ultra or Pro subscription, current eligibility and account model choices

A developer API key belongs to the service that issued it. A Google key and a reAPI key are not interchangeable. On reAPI, create your key in API key settings, then follow the endpoint and authentication examples in the model's documentation.

Google's October 13 announcement distinguishes the products: developers use the Gemini API paid tier, while Google AI Ultra and Pro subscribers can use Gemini 4 Argon in the Gemini app and AI Mode in Search^[3]. The developer guide's Free-tier and Pro-plan exclusion must not be read as a blanket statement about the consumer app. A consumer subscription does not pay for reAPI requests.

What changed for developers

The September announcement named paid API customers and Google AI Ultra subscribers as the next groups after early testing^[2]. The current guide now supplies the missing developer pieces: the exact model ID, working request shapes, thinking settings and migration guidance^[1]. It is no longer accurate to tell developers that Google has published no API.

The October 13 post explicitly announces broader global availability, including the API in AI Studio and Gemini platform, Google Antigravity for Ultra subscribers, and consumer Gemini app and AI Mode access for Ultra and Pro subscribers^[3]. These are distinct routes with their own eligibility and billing. This guide checked the sources on October 14, 2026.

The useful next step is a small evaluation with your own workload. Keep the same task, source material and success criteria across models. Record the thinking level, generated token usage and finish reason, so a response stopped by its budget is not mistaken for a finished answer. The benchmark comparison explains what Google's launch results can and cannot establish.

Why the initial rollout used Fairwind

In the September 30 announcement, Google named four areas it wanted to strengthen before broader availability^[2]:

Misuse. The model is designed to refuse cyber and CBRN attack requests while keeping legitimate dual-use research possible. Google says it is improving how it monitors the model's internal activations to spot misuse.
Prompt injection. Google calls Argon its most resilient model yet against indirect prompt injection and says it leads Gray Swan's Indirect Prompt Injection benchmark.
Misalignment. Monitors watch the chain of thought and actions and can stop execution when the model goes beyond what the user intended.
Hardened systems. Sandboxed environments are isolated and sealed before high-risk training or evaluation.

At the time, Google also described participation in the U.S. government's voluntary process for pre-release model access^[2]. That statement explains the initial release context; it does not override the current developer guide's availability statement.

The strongest reason is in the benchmark table. Argon ties for first on CWE-bench v1 at 68.0%, and Google reports clear gains over its 3.8 Flash Cyber model in vulnerability discovery, including on an internal benchmark spanning 20 programming languages^[2]. A model that good at finding flaws is also good at finding them for the wrong people. Google described these concerns as reasons for its phased release.

What the model limits mean when you start

Gemini 4 Argon uses one shared context of 1,048,576 tokens for input, thinking and output^[4]. A large prompt leaves less room for reasoning and the final answer. The older shorthand “1M output in one response” misses both the shared limit and the transport limits.

A non-streaming GenerateContent request returns up to 32,768 tokens; one streaming request returns up to 65,536. Longer generations use continuation across requests. Google documents automatic continuation in recent versions of its own SDKs, while REST clients must manage the loop themselves^[4]. reAPI does not silently continue a request on the server: if you need another segment, your client sends it explicitly.

The three native thinking levels are low, medium and high, with medium as the default. Thinking cannot be fully disabled for this model^[4]. Start with the default, then compare levels on the same tasks instead of assuming the largest budget is always the right choice.

Google documents hosted access rather than a downloadable model in the pages cited here. Its developer documentation also does not establish that a separate Argon app or a third-party APK is an official release. Follow links from Google's own product pages or use the reAPI request reference for this service.

FAQ

Is Gemini 4 Argon out?

Yes. Google's current guide says Gemini 4 Argon is available through the Gemini API and Google AI Studio^[1]. The initial September announcement described a more limited rollout; it is no longer the complete access picture.

What is the Gemini 4 Argon release date?

Google announced the model on September 30, 2026^[2]. Google announced broader global availability on October 13, 2026^[3]. October 14 is the date this guide rechecked those sources.

What is the Gemini 4 Argon model ID?

The official model ID is gemini-4-argon^[5]. Keep this identifier unchanged in requests; the display name with spaces is for readers.

Can I use Gemini 4 Argon for free?

Google documents a paid developer API tier and separate consumer access for Google AI Ultra and Pro subscribers^[3]. For reAPI, check the current token rates on the model page; do not assume a free production tier.

Does developer access require Fairwind membership?

Google now documents Gemini API and AI Studio access separately from the early Fairwind rollout^[1]. Fairwind remains part of the model's security-program history; it should not be presented as the only developer entry point.

Can one request return a million tokens?

A native non-streaming request is capped at 32,768 tokens and a streaming request at 65,536. Long decode continues across requests within the shared context budget^[4]. The client must handle that continuation when using reAPI.

Is Gemini 4 Argon available in my country?

Google calls the October 13 release globally available, but the announcement does not list eligibility for every country, account and product^[3]. Check the route you intend to use and its current regional terms.

Starting an evaluation of Gemini 4 Argon

Choose a small task with an answer you can verify: a migration plan for one module, a report grounded in supplied documents, or a structured result your application can validate. Check the response and token usage before expanding the workload. Google's current guide makes an evaluation possible; your own test tells you whether Gemini 4 Argon fits the work.