Gemini 4 Argon Reaches General Availability - API Introductory Pricing Runs to November 30, Then $4 Input and $20 Output From December 1

On October 13, 2026, Google made Gemini 4 Argon (model ID gemini-4-argon) generally available ¹. When Google announced the model on September 30, only part of a group of trusted cyber defenders in the Fairwind Program had it ⁶, and our article at the time noted that neither a date for developers, enterprises and consumers nor an end date for the introductory price had been given. Thirteen days later, both are settled.

On the API, the introductory rate of $2 for input and $10 for output (per million tokens) stays in place until November 30, 2026. Once that period ends, $4 for input and $20 for output apply ¹. The Gemini API pricing page lists the same figures for the rates that start on December 1 ².

Where it is available, and on which plans

According to Google’s announcement, access is split as follows ¹.

• API: For developers and enterprises, through AI Studio and the Gemini platform. Developers use the Gemini API on the paid tier
• Google Antigravity: Google AI Ultra subscribers
• The Gemini app and AI Mode in Search: Google AI Ultra and Google AI Pro subscribers
• Fairwind Program: Google has started adding hundreds more partners. The version without cyber guardrails goes to defenders who apply, on a rolling basis

The September 30 announcement said the broad release would start with customers paying for the API and with Google AI Ultra subscribers ⁶. This time, AI Pro subscribers are included for the Gemini app and AI Mode. Google has announced that model access in the Gemini app is being divided by plan from October, but for Argon the announcement names only AI Ultra and AI Pro and says nothing about AI Plus or users without a plan.

“AI Pro” does not mean the same thing everywhere, though. The Gemini API pricing page and the model migration guide both carry a note saying Argon cannot be used with the Free tier or with a Google AI Pro plan ²³. The page explaining Google AI plans in AI Studio says the developer benefits of those plans apply only inside the AI Studio web interface, and its table marks Argon as unavailable on AI Pro and available on AI Ultra ⁴. Read together, these pages suggest that, as a Google AI plan benefit in AI Studio, Argon is included with Ultra but not with AI Pro. The same plans page also states that calling the API directly with a key is billed separately from the subscription.

Rates double on December 1

The pricing page gives each processing option two rates: one through November 30 and one from December 1 (USD per million tokens) ².

Option	Input	Output (incl. thinking tokens)	Caching
Standard (to 11/30)	$2.00	$10.00	$0.10
Standard (from 12/1)	$4.00	$20.00	$0.20
Batch / Flex (to 11/30)	$1.00	$5.00	$0.05
Batch / Flex (from 12/1)	$2.00	$10.00	$0.10
Priority (to 11/30)	$3.60	$18.00	$0.18
Priority (from 12/1)	$7.20	$36.00	$0.36

Batch and Flex cost half of Standard, Priority costs 1.8 times Standard, and every option doubles on December 1. Every cell in the Free tier column reads “Not available,” so even trying Argon through the API means paying on the paid tier. Our previous article said the announcement did not state whether rates varied by processing option or context length. The table answers the first part, and it has no rows splitting the price by context length.

On the same pricing page, the introductory price for Gemini 3.8 Flash runs until December 31, 2026 ², so Argon’s introductory period ends a month earlier. A footnote in Google’s announcement limits this pricing to the publicly released Argon model ¹. It gives no price for the version without guardrails that goes to Fairwind participants.

Specs to check before integrating

A single context window of 1,048,576 tokens is shared by input, thinking and output ³. Our previous article followed the September 30 announcement in saying the output limit was going from 64K to 1M tokens, but the migration guide shows that reaching 1M comes with conditions.

The default maximum output is 65,536 tokens. Output beyond 64K is called “long decode,” and it becomes available when background execution is turned on and max_output_tokens is set above 65,536. The model writes in segments of at most 64K, each picking up where the last one stopped. Background execution is what runs this continuation on the server side, and if it is turned on without a max_output_tokens value, generation stops at 64K ³. Google’s announcement calls this mechanism the “continuation protocol” ¹. Clients that want to control the continuation themselves can do so with the GenerateContent API, and support in the Interactions API is listed as coming soon.

The migration guide’s checklist covers the points most likely to affect existing code ³.

• Explicit caching is not supported. Requests that create cachedContents or reference cached_content need to be removed. Implicit caching switches on automatically once a prompt reaches 4,096 tokens, and cache hits get 95% off the input price
• temperature, top_p and top_k are ignored. Setting them has no effect, and the guide says to remove them
• A max_output_tokens of 32,768 or higher is recommended, because Argon devotes a larger share of that budget to thinking compared with earlier models. Thinking depth can be set to low, medium or high, with medium as the default
• Without background execution, HTTP client timeouts should be at least 15 minutes, since difficult requests may run for several minutes

On the pricing page, thinking tokens are billed at the output rate ². Because this model puts more of the budget into thinking, it seems safer to measure output usage at each thinking level on your own workload before building an estimate.

In GitHub Copilot: Pro+, Max, Business and Enterprise

GitHub also made the model generally available in Copilot on October 13 ⁵. Copilot Pro+, Max, Business and Enterprise users get it. Visual Studio Code, Visual Studio, Copilot CLI, Copilot coding agent, the GitHub Copilot app, JetBrains IDEs, Xcode and Eclipse all list it as a choice in their model pickers. The rollout is gradual.

GitHub charges for it through usage-based billing at the provider’s list price, and leaves the actual rates to Copilot’s models and pricing page ⁵. GitHub reports that in its early tests Argon completed more tasks than earlier Gemini models across all of its coding benchmarks, and that Copilot CLI showed the largest improvement.

Organizations should note the default behavior. Business and Enterprise administrators can manage access to Argon with the model policy setting in Copilot, but under default model enablement a new model is switched on automatically unless the administrator has turned the global default off or disabled this model explicitly ⁵. Organizations running the global model policy, made generally available in August, with the default left on will see usage-billed Argon appear in their users’ model pickers. To keep it out, an administrator has to disable the model explicitly.

December 1 is the baseline for estimates

In our previous article, we wrote that Argon would stay off the shortlist for now because neither the availability date nor the date of the price increase was known, so the figures could not be used for estimates. Now they can. The introductory price, however, covers just under seven weeks, from October 13 to November 30.

Evaluation and prototyping can be done within that window, but production budgets need to be built on the December 1 rates of $4 input and $20 output. Batch or Flex brings that down to $2 input and $10 output even after December, which matches the current Standard introductory price. Code that relies on explicit caching will need its calls rewritten, and sampling settings such as temperature are ignored, so swapping the model ID alone may not reproduce the same behavior. Which of these matters depends on whether you reach Argon through the API, the Gemini app or Copilot; each has its own plan requirements and its own billing.

Sources

1. Gemini 4 Argon: Built to solve your hardest problems - Google official blog (October 13, 2026)
2. Gemini Developer API pricing - Google AI for Developers (Gemini 4 Argon pricing table, checked October 14, 2026)
3. What’s new in Gemini 4 Argon - Google AI for Developers (specs and migration guide, checked October 14, 2026)
4. Google AI plans - Google AI for Developers (Google AI plan benefits in AI Studio, checked October 14, 2026)
5. Gemini 4 Argon in GitHub Copilot - GitHub official changelog (October 13, 2026)
6. Gemini 4 Argon: our next era of frontier intelligence - Google official blog (September 30, 2026 announcement)