Gemini 4 Argon opens wider, with different access rules for each product

Google has widened access to Gemini 4 Argon, including Google AI Pro and Ultra subscribers in the Gemini app and paid developers through the API. A Pro subscription does not provide the same Argon entitlement everywhere: Google's October 13 launch post specifies Ultra for Antigravity access.

The change follows the September 30 restricted release to trusted cybersecurity defenders. Google dated its broader rollout announcement October 13, 2026, the same day as The Verge's original report, which led to this story. The earlier date is the model announcement; the later one is the expansion of access.

Choose the product before the subscription

The useful comparison is between entry points, rather than a single promise of worldwide availability. Google's launch post and Gemini app release notes describe these routes:


Where you want to use Argon	Access described by Google
Gemini app	Google AI Pro or Ultra subscribers; rollout began October 13
Google Antigravity	Google AI Ultra subscribers
Gemini API in AI Studio	Paid API tier

The API guide separately says Argon is unavailable on its free tier or Google AI Pro plan. That should be read alongside the app's explicit Pro support: access in the consumer app does not establish access in another product. Check the intended product's model selector and terms before changing a subscription or an application's configuration.

TechKili's earlier explanation of Fairwind access covers the restricted first stage. The new rollout gives a broader audience a route to evaluate the model; it does not establish identical features or entitlements across those routes.

For application builders, the API documentation describes a unified window of 1,048,576 tokens. Input, thinking and output consume that same window. A large document therefore leaves less room for subsequent reasoning and the answer; the headline capacity is not a promise of a million output tokens after an arbitrarily large input.

The guide also distinguishes ordinary responses from long output. Background execution with an output limit above 65,536 enables continuation across smaller generation steps. Developers need to account for completion status and assemble the resulting output steps. These are integration requirements, not measures of answer quality.

Evaluate a task you can check

Google presents Argon as suited to demanding professional work. That positioning alone cannot tell a team whether it improves its own workflow. TechKili's recommendation is to start with a representative task whose result can be checked: a code change with existing tests, for example, or a document summary checked against the original.

Record correctness, completion time and usage together. Keep approval for consequential actions in the surrounding workflow. This article assesses published access and integration information; it does not report independent performance testing.