Gemini 4 Argon Now Generally Available: Plans, API Pricing and What Changed Since the Announcement | CoAI

On October 13, 2026 (U.S. time), Google made its new frontier model Gemini 4 Argon generally available. At the time of the September 30 announcement, it was offered only to a limited number of organizations working in cyber defense, but it is now available to developers, businesses and general paid-plan subscribers. In its official blog, Google said it is "rolling out globally starting today."

Where it is available and on which plans

According to Google, Gemini 4 Argon is offered in the following places.

For developers and businesses: the paid tier of the Gemini API, the web, desktop and mobile versions of Gemini for businesses (Gemini for business), and Google Cloud
Google AI Ultra and Google AI Pro subscribers: the Gemini app and AI Mode in Google Search
Google AI Ultra subscribers: the agentic development environment Google Antigravity

The September 30 announcement said general availability would start with Google AI Ultra subscribers, but Google AI Pro subscribers can now also use it in the Gemini app and AI Mode in Search. In the Gemini app, "4 Argon" is added to the model options. All of these are gradual rollouts, so it will not appear on everyone's screen right away.

On the other hand, availability on the free plan or Google AI Plus has not been announced. The Gemini API pricing page clearly states that it cannot be used on the free tier or with a Google AI Pro plan. In the developer tool Google AI Studio, only Ultra gives access through a subscription plan; Pro subscribers can use it in the Gemini app and AI Mode in Search.

Availability in Japan

Google's announcement describes the availability as "global" and does not set conditions by country or language. ITmedia reported that subscribers in Japan to plans such as Google AI Pro (2,900 yen per month) can also use it, and that it confirmed Argon can be selected on the Pro screen. However, it said it is unclear when Google Workspace users will be able to use it. Which languages AI Mode in Search supports has not been officially disclosed. The integrations with more than 40 third-party apps announced for the Gemini app at the same time are mainly for English in the U.S., with some also said to be available in other English-speaking countries, Brazil and India.

Specifications for developers

The model ID in the Gemini API is gemini-4-argon, and it is offered as a stable version. The main specifications are as follows.

Up to about 1 million tokens (1,048,576 tokens) of input and output combined
Output is up to 65,536 tokens by default, and up to about 1 million tokens when using the setting for generating long outputs
Input can be text, images, video, audio and PDF. Output is text only
Three thinking levels: low, medium and high
Supports function calling, code execution, and grounding with Google Search and Google Maps. Computer use is in preview
Knowledge cutoff is March 2026

It does not support image or audio generation, the Live API for real-time interaction, or explicit context caching. Only implicit caching, which is applied automatically, is available.

Pricing

Pricing is the same as in the September 30 announcement, but the end date for introductory pricing has now been made clear. According to the Gemini API pricing page, prices per million tokens are as follows.

Until November 30: $2 for input, $10 for output (including thinking tokens), $0.10 for cached input
From December 1: $4 for input, $20 for output, $0.20 for cached input

For Batch and Flex, which send non-urgent jobs together, prices are $1 for input and $5 for output until November 30, doubling from December 1.

Changes since the announcement and safety

Compared with the September 30 announcement, the main differences are as follows.

Google AI Pro has been added alongside Google AI Ultra to those eligible for general availability
November 30 has been given as the end date for introductory pricing
Regarding the "1 million output tokens" mentioned at the announcement, it has become clear that the standard output limit is 65,536 tokens, and producing up to about 1 million tokens requires the long-output setting (a continuation mechanism)

Google said the Fairwind Program for cyber defense professionals is now starting to expand to several hundred additional organizations. In this program, unlike the general version, participants can use a version with cyber-related safeguards removed for defensive purposes. The general version of the model has cyber-related safety measures built in by default. In its model card, Google says it cannot rule out that the model may have reached the "critical capability levels" it regards as dangerous in cybersecurity and biology, and that it is offering the model with mitigations in place.

On performance, as Google's own announcement, it said the attack success rate on Gray Swan's benchmark, which measures resistance to indirect prompt injection, was held to 0.7% over 15 adaptive attacks, and the model card gives figures such as 19.6% on the legal agent evaluation Harvey's Legal Agent and 84.2% on the 256,000 to 1 million token segment of GraphWalks, which tests handling long contexts. The figures from the announcement of 77.9% on DeepSWE v1.1 and 68% on CWE-bench v1 have not changed. In addition, Google said it will provide up to 10,000 researchers, selected from applicants, with six months of free access to the top usage tier of Google Antigravity (conditions such as eligibility and capacity apply).

CoAI's take

In less than two weeks after the announcement, it has become available even to Google AI Pro subscribers. Compared with the original explanation that it would start with Ultra subscribers, it is fair to say it reached the general public faster. For developers, the period until November 30 is a window to try it at introductory pricing. Since prices double from December, for workloads that use long outputs heavily, it is a good idea to check the costs of your own usage during this period. Meanwhile, there is still no official explanation on points that matter to users in Japan, such as AI Mode in Search in Japanese or availability in Workspace.

Sources

Tools mentioned in this article

GeminiGoogle's conversational AI, integrated with Search and Workspace.View details

Back to news

Gemini 4 Argon Is Now Generally Available, With the Gemini App Open to AI Pro and Ultra and Introductory API Pricing Running Until November 30