Gemini 4 Argon: Google opens its model to everyone at $2 per million input tokens

For the first time, Google has opened Argon in the paid API, in Antigravity and in the Gemini app. Anyone who pays can now put it inside their own agent.

At the Sep 30 announcement, Argon went to cyber defenders in the Fairwind program, internal Google teams and trusted testers, and there was no public API at all. On Oct 13, four doors opened: the paid Gemini API in AI Studio, Google Antigravity, the Gemini app and AI Mode in Search.

Introductory pricing. $2 per million input tokens and $10 per million output tokens. On Dec 1, 2026, Google will raise them to $4 and $20, and cached input is billed at a 95% discount.

How to connect. The model code is gemini-4-argon, with a 1,048,576-token context as a single limit for input and output. In normal mode the model returns up to 65,536 tokens at a time, and up to a million in long decode mode. The models.dev registry already lists the model under the google provider, and opencode pulls it from there.

You will need a paid key. On Oct 13, the request to support Argon in the Antigravity OAuth plugin for opencode was closed as not planned, and the model is not in Antigravity's server catalog.

Subscriptions unlock different things: Antigravity and AI Studio work on AI Ultra at $99 a month, while the Gemini app and Canvas also work on AI Pro at $20.

On coding, the picture is uneven. On DeepSWE v1.1, Argon scores 77.9% and beats Claude Opus 5.5 by 3.7 points. On Terminal-Bench 4.0, it trails the same Opus 5.5 by 9 points, and on FrontierSWE v2, at 55.0%, it is 10.5 points behind GPT-6 Astra. In the composite Vals Index, Argon is first with 68.9% and leads in 13 of 19 rows.

On Dec 1, 2026, the introductory pricing ends: input will cost $4 per million tokens and output $20.