Gemini 4 Release Date: Confirmed, Not Released

Updated October 30, 2026: Gemini 4 Argon has been callable on Google's paid Gemini API since October 13, 2026, and Gemini 4 Carbon was announced on October 28, 2026; see our Gemini 4 Argon review and Gemini 4 Carbon: release date, price and API access.

The Gemini 4 release date came in two steps. Google announced Gemini 4 Argon on September 30, 2026 and opened it to developers on the paid Gemini API tier on October 13, 2026. Google then announced Gemini 4 Carbon on October 28, 2026, with no developer date. Google's pricing page lists Argon as gemini-4-argon at $2 per million input tokens and $10 per million output tokens through November 30, 2026, then $4 and $20 from December 1.

The rest of the family is still open. No Gemini 4 Pro or Gemini 4 Flash has been announced, and Google's Carbon post names no model ID, no endpoint and no developer date.

Below are the official statements in order, the dates that hold, the Gemini 4 model you can call today, and how to check your own model list before moving production traffic. We did not run or claim a Gemini 4 benchmark; every figure below traces back to one of those pages.

Executive answer

What Google actually confirmed

Google named Gemini 4 once in its 2026-07-21 announcement for Gemini 3.6 Flash, 3.5 Flash-Lite, and 3.5 Flash Cyber. The relevant sentence is short:

“We have started our most ambitious pre-training run yet, for Gemini 4, and are excited by the progress.”

That July sentence said three things:

1. Google uses the name Gemini 4.
2. A pre-training run had started by 2026-07-21.
3. Google called it its most ambitious pre-training run at that time.

It said nothing about a completion date, a model ID, tier names, context window, modalities, benchmark results, price, region, quota or API access. Each of those needed its own announcement, and the September 30 post supplied several of them.

The two posts are also written differently. The July post described Gemini 3.6 Flash and 3.5 Flash-Lite as products developers could use, and said Gemini 3.5 Pro was testing with partners. Gemini 4 appeared only in the forward-looking sentence about the next generation, while the September 30 post led with a name, a price and a rollout cohort.

Why pre-training is not a release

Three different states get collapsed in Gemini 4 headlines, and Argon shows what each one looks like:

Gemini 4 sat in the first state from July 21 to September 30, 2026, and moved through the second to the third on October 13, 2026. Reading a pre-training run as a launch creates broken configuration, invented budgets and comparisons against a model nobody can call.

The Gemini consumer app and the Gemini API are separate products too. A name in an app picker, an experiment, an arena, or a partner preview would not by itself create an API model ID.

Is there a Gemini 4 release date?

Yes, and it has two dates. Google announced Gemini 4 Argon on September 30, 2026 and said developers could start building with it on October 13, 2026 in Google AI Studio and Android Studio, on the paid Gemini API tier. Gemini 4 Carbon was announced on October 28, 2026, and Google gives it no developer date: the announcement says Carbon will reach developers, businesses and consumers “as soon as possible”.

Late-2026 and October forecasts filled search results and prediction markets before the launch, and Argon landed on September 30. A forecast that lands on the right month still leaves developers without a model ID, a rate card and the supported parameters, so check Google's own pages before you plan.

Watch these five pages instead of a rumor calendar:

1. Google's Gemini API model catalog.
2. The official Gemini API changelog.
3. The Gemini API pricing table.
4. The Models API response for your own account and region.
5. A model page or migration guide that documents the exact ID and parameters.

Until at least the first four agree, do not put a Gemini 4 identifier into production configuration.

Gemini 4 Pro, Flash, pricing, and specs: what is announced

Autocomplete demand makes the Pro and Flash names sound settled. Google has announced two Gemini 4 models, Argon and Carbon, and neither uses those tier names.

Google has used Pro, Flash and Flash-Lite elsewhere. That naming history makes the guesses understandable, but a naming pattern is not an API spec. The rest of the family could use those names, different names, or a staged release that exposes only one variant first.

The same rule applies to inherited specifications. Do not copy Gemini 3.8 Flash's 1,048,576-token input limit, 65,536-token output limit, thinking levels or rates into a Gemini 4 spreadsheet. They describe a different model.

Rumor log: what held up and what did not

The important pattern is independence. Ten articles that cite the same anonymous claim are one source, not ten confirmations. A screenshot without a request, response, account scope, and exact endpoint is a lead to investigate, not a specification to publish.

Check Gemini 4 availability yourself

Google's Models API lists the models available to your Gemini API credential. Keep the key in an environment variable and send it in the x-goog-api-key header rather than placing it in a URL:

set -o pipefail
curl --fail-with-body --silent --show-error \
  https://generativelanguage.googleapis.com/v1beta/models \
  -H "x-goog-api-key: $GEMINI_API_KEY" \
| jq -r '.models[]?.name | sub("^models/"; "") | select(startswith("gemini-4"))'

Any gemini-4-* line in the output is a Gemini 4 model that credential can call. An empty result means that account cannot call any gemini-4-* model yet, so do not configure it as though one were available.

Check the gateway you actually plan to call as a separate step:

set -o pipefail
curl --fail-with-body --silent --show-error https://api.omniakey.com/v1/models \
  -H "Authorization: Bearer $OMNIAKEY_API_KEY" \
| jq -r '.data[]?.id | select(startswith("gemini-4"))'

Provider and gateway lists answer different questions. Google's response shows what Google exposes to that credential; OmniaKey's response shows what OmniaKey serves right now. A model can appear on one list before the other.

Use exact matching. A response mentioning “Gemini 4” in descriptive text, or an ID such as gemma-4-*, is not a gemini-4-* model.

Which Gemini 4 model you can call today

Two Gemini 4 models have been announced, and one of them is callable: Gemini 4 Argon (gemini-4-argon), generally available on the paid Gemini API tier since October 13, 2026. Carbon was announced on October 28, 2026, with access starting with a subset of cyber defenders and no developer date.

The Gemini 4 Argon model page shows the current price.

Gemini 3.8 Flash stays the cheaper option for general text, coding and tool-using agent work:

Those rows are Google's numbers, not Gemini 4 predictions. The Gemini 3.8 Flash pricing guide separates Google Standard, Batch, Flex and Priority, then works out cost per task that passes. The Gemini 3.8 Flash model page shows the current OmniaKey price for that model.

One replacement is already scheduled: from December 1, 2026, Google sends every gemini-2.5-pro request to gemini-3.8-flash, and the 2.5 Pro deprecation and migration guide lists the request changes that come with the reroute.

When you need the strongest Google model, Argon answers that requirement today. Run your own tasks against Argon and Gemini 3.8 Flash with a fixed test set, then compare cost per task that passes, latency, tool reliability and failure types instead of launch adjectives.

Prepare for the rest of Gemini 4 without guessing IDs

You can make the next migration cheap while the family is still partial:

1. Keep the model ID in runtime configuration, but keep its current value valid.
2. Add an exact model-list preflight to deployment smoke tests.
3. Freeze a representative test set and the rubric that decides whether a result is good enough.
4. Record input, cached input, thinking, visible output, retries, tools, latency, and final cost.
5. Isolate Gemini-native fields from OpenAI-compatible fields so a new parameter set does not leak across clients.
6. Require an explicit rollout decision after the model page and rate card exist; do not auto-switch on a name sighting.

This turns launch day into a measured access decision. Without the baseline, even a real Gemini 4 benchmark tells you little about your prompts, tools, permissions, and retry policy.

What will count as a Gemini 4 release

Argon is the working example. A Gemini 4 model counts as released here when five things exist:

1. A Google model page that names the exact ID.
2. A dated changelog entry or launch announcement.
3. A pricing row or an explicit free-preview period.
4. A Models API response that exposes the ID to an eligible account.
5. Google's published limits, supported methods, regions and lifecycle state.

For Argon, Google's own pages cover the list: the model catalog lists gemini-4-argon as Stable, the October 13, 2026 changelog entry marks it generally available, and the pricing page carries its rates. Run the Models API check above to confirm your own account can call it before you pin it in production.

OmniaKey availability is a separate check. Google releasing a model does not mean every gateway carries it on day one, and OmniaKey serves Argon today.

Common mistakes

Treating an official name as an official product

Gemini 4 is officially named, and Argon and Carbon are the two named models. A third name in a headline still needs its own announcement.

Turning family history into model IDs

gemini-4-pro and gemini-4-flash look reasonable because older families use those tiers. Sending either string today is still a guess.

Mixing the Gemini app with the Gemini API

Consumer plan access, an experimental picker, Google AI Studio, Vertex AI, and the Gemini API can move on different schedules. Always name the product behind an access claim.

Publishing benchmark comparisons before access exists

Google's Argon table is the vendor's own run, and no reproducible public benchmark covers your prompts. Scores without prompts, settings, model identity, and artifacts are not a review.

Budgeting with inherited prices

Gemini 3.8 prices are a current baseline, not a forecast for Carbon. Carbon's announced rate matches Argon's introductory rate, and Google has not said how long Carbon's introductory period lasts.

Frequently asked questions

Has Gemini 4 been released?

Partly. Gemini 4 Argon (gemini-4-argon) has been generally available on Google's paid Gemini API tier since October 13, 2026. Gemini 4 Carbon was announced on October 28, 2026 and is not open to developers yet.

When is the Gemini 4 release date?

Argon: announced September 30, 2026, generally available from October 13, 2026. Carbon: announced October 28, 2026 with no developer date; Google says it will reach developers “as soon as possible”.

Is Gemini 4 Pro confirmed?

No. Google has announced Gemini 4 Argon and Gemini 4 Carbon, and neither uses the Pro tier name. No gemini-4-pro ID has been announced.

How much will the Gemini 4 API cost?

Argon costs $2 per 1M input tokens and $10 per 1M output tokens through November 30, 2026, then $4 and $20 from December 1, 2026. Google says Carbon will launch at the same introductory rate, with no date for when that rate starts.

Does OmniaKey support Gemini 4?

Yes for Argon: the Gemini 4 Argon model page shows the current price. Carbon has no model ID yet, so there is nothing to configure.

Official sources and rumor audit

Official sources, checked on 2026-10-30:

• Google: Introducing Gemini 4 Argon — the September 30, 2026 launch and the Argon prices
• Google: Gemini 4 Argon is now available for developers — the October 13, 2026 paid API access
• Google: Introducing Gemini 4 Carbon — the October 28, 2026 announcement, with no developer date
• Google: Gemini 3.6 Flash, 3.5 Flash-Lite, and 3.5 Flash Cyber announcement — the primary Gemini 4 pre-training statement
• Google: Gemini API model catalog
• Google: Gemini API changelog
• Google: Gemini API pricing
• Google: Models API reference
• Google: Gemini 3.8 Flash model page
• OmniaKey live model catalog and the Gemini 4 Argon model page

Representative secondary coverage was read only to trace the rumors: Fello AI's status overview and MindStudio's release-rumor article. Their forecasts and unattributed technical claims are not used as product facts here.

Sources were checked on 2026-10-30. Release state, model IDs, prices, and limits can change; re-run the model-list checks and read Google's rate card before changing production configuration.