Gemini 4 Argon Access: Pro, Ultra, API and Antigravity
Find the right Gemini 4 Argon entry point, distinguish App and API access, and diagnose missing models without assuming a subscription unlocks every product.
Ofox 
Gemini 4 Argon is available through several Google products, but the eligibility rules differ by product. The Gemini app announcement includes Google AI Pro and Ultra. Antigravity’s model table includes Ultra and Enterprise, and excludes the old standalone Antigravity IDE. For developers, Google identifies the paid Gemini API tier; an app subscription is not evidence that a particular API project can call the model.
This guide reconciles the official announcements and documentation checked on October 14, 2026. It is a documentation-based access guide, not an account activation test. Use the matrix to identify your route, then check the account and client that will actually run the task.
Where can you use Gemini 4 Argon?
| Entry point | What the official sources establish | What you still need to check |
|---|---|---|
| Gemini app | Google AI Pro and Ultra subscribers are included in the October 13 announcement | The signed-in account, its subscription, and the model options shown to that account |
| Google AI Studio | The developer announcement specifically includes Ultra subscribers trying Argon | Which account/project is selected and whether you are using the Studio interface or a billed API project |
| Gemini Developer API | The announcement names the paid API tier; the model ID is gemini-4-argon | Project billing, model access and applicable quota; do not infer these from an app subscription |
| Antigravity app, CLI and IDE extensions | The model table lists Ultra and Enterprise for Argon | Correct subscription/organization, supported client and selected model |
| Old standalone Antigravity IDE | Explicitly excluded by the Argon footnote | Moving to a supported client is necessary; refreshing this IDE is not a documented workaround |
| Ofox API | Our October 14 public catalog and authenticated model-list checks did not list Argon | No verified Ofox Argon route at the time of this check; do not guess a provider-prefixed ID |
Sources: Google’s Gemini app announcement, developer announcement, Antigravity model table, and API model specification.
The most consequential distinction is Pro in the Gemini app versus Pro in Antigravity. These are not interchangeable access promises. Antigravity’s table marks Argon unavailable on Free/Plus and Pro, while listing it for Ultra and Enterprise. Buying or already owning Pro therefore does not establish Antigravity Argon access. Likewise, the API pricing page’s Free/Pro exclusion must be read in that API context; it does not undo the app announcement.
Choose a route based on the deliverable
For an interactive question or document discussion, start with the Gemini app route if your account qualifies. For work inside a code repository, check the supported Antigravity clients and plan matrix. For a repeatable application that must save usage, retries and results, investigate the Developer API route and its billing separately.
Before changing plans, write down three items: the product you intend to use, the signed-in account, and the deliverable you need. “I have Gemini Pro” is incomplete troubleshooting information. “The Ultra account shown in my Antigravity CLI cannot select Argon” identifies a different problem from “My Pro account cannot see Argon in the old IDE.”
Our Antigravity platform guide explains the platform background. This article focuses specifically on Argon’s access boundaries.
Argon is missing: diagnose in this order
1. Confirm the product and the exact model
Record whether you are in the Gemini app, AI Studio, an API application, the Antigravity app, CLI, extension or old IDE. For API integration, use the documented ID gemini-4-argon; a marketing family name such as “Gemini 4” is not a substitute for an API identifier.
Expected result: you can point to one row of the matrix. If you cannot, first identify the client rather than changing billing or copying credentials between products.
2. Match the active account to its entitlement
Check the account displayed in the client against the account carrying the subscription or enterprise membership. Where a project or organization selector is present, record that too. A subscription visible in one browser account does not prove a CLI is authenticated as the same user.
Expected result: an account/product/plan combination supported by the official matrix. If the plan is excluded, stop treating the issue as a hidden setting. Review the supported route before making a purchase; this guide does not establish country-specific purchase or account eligibility.
3. Check the client boundary
For Antigravity, the footnote is decisive: Argon is available in the app, CLI and IDE extensions, not the old standalone IDE. Record the actual application and version before filing a report. The similar names make this easy to miss.
Expected result: a supported client with its model selector checked. Do not assume installing an extension retroactively adds Argon to every Antigravity interface.
4. Separate visibility from a successful task
Seeing Argon in a selector establishes visibility, not that a task finished or that unlimited usage is included. A request can still encounter quota, billing, permissions or execution errors. Save the actual error and timestamp before retrying. For an API request, retain the request ID and response status if provided, but remove credentials and private prompt content from any report.
We have not run a paid Argon request for this article. No account-specific success, response time or quota is claimed. For cost planning before your first call, use our Argon API pricing worksheet.
5. Escalate with a small, reproducible report
Use this template for your own notes or the appropriate product support channel. It is a diagnostic checklist, not a guaranteed fix:
Checked at (with time zone):
Product and client version:
Model requested or selector label:
Account category (personal/work; redact identity):
Plan / project billing status:
Organization or project selected (redacted):
Official eligibility row checked:
Observed behavior:
Error code and request ID, if supplied:
Smallest reproduction without confidential content:A complete report lets a reviewer distinguish an unsupported route from an account-specific failure. Repeatedly reinstalling a client or upgrading subscriptions before making this distinction can leave the original cause unresolved.
What Argon can accept—and what it does not generate
The API specification lists text, image, video, audio and PDF inputs, with text output. Image generation, audio generation and Live API are not supported by this model. An interface may combine several models or tools; that does not mean the Argon API itself generates an image or an audio track.
The documented context window is 1,048,576 tokens shared by input and output. The default synchronous maximum output is 65,536 tokens. Longer decode uses a separate workflow and the remaining context budget; a one-million-token context is not a promise of one million output tokens on top of a full input. Implicit caching is supported, explicit caching is not. These limits matter when deciding whether access alone will satisfy your task. See the model specification and long-decode documentation.
Is October 13 the first time anyone had access?
No. Artificial Analysis’s September 30 analysis discussed restricted early access. Google’s October 13 announcements describe broader product availability. These statements concern different dates and scopes. An older “selected users only” report should not override a later product access table, and a broad announcement should not be read as proof that an individual account is enabled.
Access checklist before you spend money
You should be able to identify the product, account, qualifying plan or billed project, supported client and exact model. Then confirm a small task completes, inspect the result, and capture usage where available. Only after that should you scale the workflow.
Access also does not establish which model will solve your task best. Our Argon benchmark comparison separates coding, terminal and long-context evidence so you can choose a meaningful pilot rather than treating availability as a performance verdict.
