A blue glass star above three gateways illustrating Gemini 4 Argon development and future availability milestones

model-release

Gemini 4 Argon Release Date: Who Can Use It?

EvoLink Team

Product Team

September 16, 2026

Updated on October 13, 2026

14 min read

Google announced Gemini 4 Argon on September 30, 2026. Its announcement described an initial Fairwind rollout and plans to expand access to paid API customers and Google AI Ultra subscribers, without a public rollout date. Those Google milestones are separate from gateway access: Gemini 4 Argon is available through EvoLink; use the model page for the current integration example and pricing.

For teams using EvoLink, the next step is to evaluate Argon on a representative task while keeping the current model as a fallback. This timeline explains what Google announced and which channel each availability statement describes; the comparison guides help you decide which workload to test.

Gemini 4 Argon evaluation milestones: announcement, API contract, route verification and workload evaluation

"Release Date" Can Mean Four Different Events

Announcement, restricted testing, consumer access and callable API access are separate milestones. The announcement and initial Fairwind rollout are confirmed; the remaining channels need their own evidence.

Milestone	What would count as confirmation	Dated evidence and channel status	What would not count
Official announcement	A Google or Google DeepMind post naming the model, its variants, and launch scope	Google announced Gemini 4 Argon on September 30, 2026.	Earnings-call phrasing, anonymous leaderboard checkpoints, prediction articles
Gemini app availability	Google states that Gemini 4 is live in the Gemini app, with plan or region scope	Google's announcement names Google AI Ultra subscribers in the planned expansion, without dating a public Gemini app rollout. Fairwind access is a separate restricted channel, not proof of app availability.	Screenshots of unlabeled or A/B-tested models
Gemini API / AI Studio / Vertex availability	A catalog entry with model ID, pricing row, rate limits, and changelog line on ai.google.dev or Vertex docs	No Gemini 4 Argon entry or request model ID was found in the public Gemini API catalog or changelog when checked on October 1, 2026.	An app launch, a keynote demo, or a third-party gateway listing
Gateway (EvoLink) availability	Approved model ID, price, a successful authenticated request with matching returned identity, and reconciled usage and billing	Gemini 4 Argon is available through EvoLink. See the model page for current pricing and the request example.	A Google announcement, or a placeholder page carrying the model's name

A model can be announced before the app receives it, reach the app before the API lists it, and be listed upstream before any gateway has verified a route. These are separately evidenced states; one channel’s launch does not establish access through another.

The Direct Answer: What Is Confirmed Today?

Question	Answer and evidence scope	Evidence level
Is Gemini 4 released?	Announced · Restricted rollout	Google
Has Google confirmed Gemini 4 exists?	Google announced Gemini 4 Argon on September 30, 2026.	Google
Public API release date	The public API release date has not been announced.	Google
Model variants (Pro, Flash, Flash-Lite)	Gemini 4 Argon is the announced name. Pro, Flash and Flash-Lite variants are not confirmed by this announcement; a product name is not a request model ID.	Google
Official model ID	No Gemini 4 Argon entry or request model ID was found in the public Gemini API catalog or changelog when checked on October 1, 2026.	Unknown
Pricing	Google's announced introductory rates are $2 per 1M input tokens and $10 per 1M output tokens. After the introductory period they become $4 and $20 respectively. The end date is not announced. These are Google rates, not an EvoLink quote.	Google
Context window, output limit, modalities, tool and thinking controls	Google announces an output limit of 1M tokens for extended reasoning and generation. This is an output limit, not a confirmed input context window or an EvoLink parameter guarantee. The public API model ID, input limits, supported request parameters, quotas remain channel-specific; use EvoLink’s model page and documentation for its integration contract.	Google
Public API (Gemini API, AI Studio, Vertex AI)	No Gemini 4 Argon entry or request model ID was found in the public Gemini API catalog or changelog when checked on October 1, 2026.	Official (absence, checked October 1, 2026)
EvoLink route	Gemini 4 Argon is available through EvoLink. See the model page for current pricing and the request example.	EvoLink status (separate from Google)

Google positions Argon for software engineering, enterprise knowledge work and cybersecurity defense, and publishes model evaluations. These are provider-reported results, not EvoLink tests or a guarantee for your workload.

Official Timeline

July 2026 pre-training statements remain historical milestones. The September 30 announcement is the current source for the name, rollout scope, output limit and announced prices.

Date	Event	What Google says	What it does not say
July 21, 2026	Google publishes the Gemini 3.6 Flash announcement on blog.google	Google has started its most ambitious pre-training run yet, for Gemini 4	No date, variants, specs, or API plan
July 22, 2026	Alphabet's Q2 2026 earnings message from the CEO is published on blog.google	"We have started our most ambitious pre-training run yet, for Gemini 4, and are excited by the progress we are seeing at the frontier."	No date, variants, specs, or API plan
2026-09-30	Gemini 4 Argon	Google announced Gemini 4 Argon on September 30, 2026. Initial access is limited to trusted cyber defenders through Google's Fairwind Program. Google plans broader access starting with paid API customers and Google AI Ultra subscribers, but has not given a public rollout date.	The public API release date has not been announced. The public API model ID, input limits, supported request parameters, quotas remain channel-specific; use EvoLink’s model page and documentation for its integration contract.
2026-10-01	Historical catalog check	The October 1 check did not find Argon in the public Gemini API catalog or changelog.	This dated observation does not establish current access through every channel.

What Is Confirmed, and What Still Needs Verification

The September announcement supersedes the July pre-training-only status. Keep these evidence boundaries when reading launch coverage:

Initial access is limited to trusted cyber defenders through Google's Fairwind Program. Google plans broader access starting with paid API customers and Google AI Ultra subscribers, but has not given a public rollout date.
Gemini 4 Argon is the announced name. Pro, Flash and Flash-Lite variants are not confirmed by this announcement; a product name is not a request model ID.
Google announces an output limit of 1M tokens for extended reasoning and generation. This is an output limit, not a confirmed input context window or an EvoLink parameter guarantee. The public API model ID, input limits, supported request parameters, quotas remain channel-specific; use EvoLink’s model page and documentation for its integration contract.
Google's announced introductory rates are $2 per 1M input tokens and $10 per 1M output tokens. After the introductory period they become $4 and $20 respectively. The end date is not announced. These are Google rates, not an EvoLink quote.

Three name collisions also pollute the results and should be excluded when you evaluate a claim:

Gemma 4 is Google's open-weights model family, a different product with different distribution. A Hacker News author publicly corrected a post that had confused the two.
NASA's Gemini 4 was a 1965 crewed spaceflight and has nothing to do with the model.
Anonymous LM Arena checkpoints. On July 2, 2026, media coverage speculated that an unnamed Gemini Flash checkpoint on LM Arena might be Gemini 4. Google did not confirm that; an unlabeled leaderboard entry is not evidence of a release. Third-party apps and repositories calling themselves "gemini-4" belong in the same category.

Which Access Channel Do You Need?

Google announcement and direct access. The September 30 announcement establishes the model name and initial rollout plan. It does not establish that every Gemini app plan, region, AI Studio account or Vertex project can use Argon. Check the channel you intend to use rather than treating an announcement as universal access.

EvoLink access. Open the Argon model page, use its current model ID and request example with your EvoLink key, and check the existing pricing section. The gateway integration and Google's direct rollout are separate. EvoLink access does not imply that your Google account has direct access.

Your workload. A working route is the start of an evaluation. Replay tasks, inspect usage and outputs, and compare accepted results with the model you already use. Keep the existing route available until the candidate meets your quality, cost and latency thresholds.

What Developers Can Prepare Now

Use the existing EvoLink integration to start with a bounded evaluation. These steps keep the launch from turning into an unmeasured production migration.

Keep the candidate model ID in configuration, not in code. Copy the current EvoLink ID from the model page. Keep it in reviewed routing configuration so a later version change does not require searching application code. The page slug is not a request ID.
Freeze your current Gemini baselines. Choose existing baselines such as Gemini 3.8 Flash and Gemini 3.1 Pro from the Gemini collection page. Record their cost per accepted task, latency percentiles, and failure rates on your own workload, and verify each route before relying on it in production. These are evaluation baselines; their prices and context limits are not Gemini 4 specs.
Save representative evaluation traces. Export a fixed set of real requests and accepted outputs from Gemini 3.8 Flash and Gemini 3.1 Pro, including tool calls, structured-output schemas, long-context cases, and the failures you had to retry. A new model is judged against your traces, not vendor benchmarks.
Define promotion and rollback gates in advance. Write down the thresholds a new model must clear (accuracy on your traces, cost per accepted task, P95 latency, error rate) and the conditions under which traffic goes back. Deciding before the launch removes the pressure to decide during it.
Log the full compatibility surface. Record which parameters, tool-calling formats, thinking controls, caching behavior, and response fields your code depends on today. Generational releases may change these contracts; a known dependency list turns a migration surprise into a checklist item.

Evaluate Gemini 4 Argon on EvoLink Compare current Gemini models

Which Comparison Should You Read Next?

For an OpenAI-based coding workflow, read Gemini 4 Argon vs GPT-6 Astra: Coding, Cost & Access. Compare accepted patches, protocol migration work and cost per accepted task before changing routes.

If you run long-horizon agents on Claude and want to know what a Gemini 4 evaluation would have to prove, read Gemini 4 vs Claude Fable 5.1: Long-Running Agent Choices. It frames access, context reuse, recovery, and cost per accepted task, with unverified Gemini 4 capabilities marked.

Signals We Monitor Before Updating This Page

The following changes affect how to interpret this timeline:

An official Google or Google DeepMind post that names Gemini 4 variants, launch scope, or a date.
A Gemini 4 entry in the ai.google.dev model catalog, with a model ID string.
A Gemini 4 row on the Gemini API pricing page.
A Gemini 4 line in the Gemini API changelog.
Gemini 4 availability in Vertex AI or Google AI Studio.
An EvoLink route verification: approved model ID, SKU, successful authenticated request with matching returned identity, and reconciled billing.

Use the Gemini 4 Argon page for EvoLink access and pricing; keep existing routes serving until your evaluation supports a change. Historical observations retain their check date. For current channel availability, follow the relevant provider documentation; the article URL stays fixed across updates.

FAQ

When was Gemini 4 Argon announced, and when is the public API coming?

Google announced Gemini 4 Argon on September 30, 2026. Initial access is limited to trusted cyber defenders through Google's Fairwind Program. Google plans broader access starting with paid API customers and Google AI Ultra subscribers, but has not given a public rollout date.

What does the 1M-token output announcement confirm?

Google announces an output limit of 1M tokens for extended reasoning and generation. This is an output limit, not a confirmed input context window or an EvoLink parameter guarantee. The public API model ID, input limits, supported request parameters, quotas remain channel-specific; use EvoLink’s model page and documentation for its integration contract.

What is the difference between Gemini 4 and Gemma 4?

Gemini 4 Argon belongs to Google's proprietary Gemini family and has restricted initial access. Gemma 4 is a separate open-weight family with its own distribution and terms; its releases do not establish Gemini access.

Will the Gemini app and the Gemini API get Gemini 4 at the same time?

Initial access is limited to trusted cyber defenders through Google's Fairwind Program. Google plans broader access starting with paid API customers and Google AI Ultra subscribers, but has not given a public rollout date.

Is the official name Gemini 4 Argon?

Gemini 4 Argon is the announced name. Pro, Flash and Flash-Lite variants are not confirmed by this announcement; a product name is not a request model ID.

Can I use Gemini 4 through EvoLink now?

Gemini 4 Argon is available through EvoLink. See the model page for current pricing and the request example. Use the Gemini 4 Argon page for EvoLink access and pricing; keep existing routes serving until your evaluation supports a change.

When is it worth starting a migration evaluation?

Run a workload-specific comparison through EvoLink while preserving your current baseline, acceptance criteria and fallback during replay, shadow and canary stages. API access enables the evaluation; it does not establish a quality or cost win.

Does Ultra-first access mean Pro subscribers will never get Argon?

No such conclusion follows from an initial rollout. Google’s announced access order does not establish a permanent exclusion or a Pro availability date. A Gemini subscription plan and a model named Pro are different things; check the current plan entitlement before subscribing for this model.

Where should I check Google API or AI Studio early access?

Start with Google’s announcement and the model access shown in your Google account. An announcement or catalog entry does not grant account access. Do not assume there is a public waitlist or an application form unless Google links one. EvoLink access is a separate route and does not grant Google AI Studio eligibility.

Should I replace an older Gemini model immediately?

Check the exact model ID and route against Google’s deprecation notices and your provider’s availability. Argon’s announcement alone is not a retirement notice for every older model. Replay representative tasks before switching, and choose a tested fallback that will still be available at migration time; do not use a retired route as rollback.

Sources

Updated October 13, 2026. The September 30 announcement was rechecked; the October 1 catalog observation remains historical. Google direct access, EvoLink access and workload suitability are separate questions.

All Posts

#Gemini 4 #Gemini 4 Argon #Google #Release Date #Release Watch #Gemini