Gemini 4 Argon Goes Global With a 1M-Token Limit and Paid API Access

Google has broadly released Gemini 4 Argon across its API, enterprise and consumer products, pairing a one-million-token limit with introductory paid pricing and a detailed model card.

2 min read

A developer workspace with multiple monitors, notes and a task checklist, illustrating software workflow management.

A developer workspace with multiple monitors, notes and a task checklist, illustrating software workflow management.

Illustrative photo by Jakub Żerdzicki on Unsplash; not a Google product image.

Google made Gemini 4 Argon broadly available on October 13, moving its frontier model beyond the limited rollout announced in September. Developers and enterprises can access the model through the paid Gemini API tier in AI Studio and the Gemini platform. Google AI Ultra subscribers can use it in Antigravity, while Ultra and Pro subscribers can access it in the Gemini app and AI Mode in Search.

The accompanying model card describes Argon as a natively multimodal, sparse mixture-of-experts model accepting text, images, audio and video. Its unified input-output token limit reaches one million tokens, although outputs are text. Google lists a March 2026 knowledge cutoff and warns that the model can hallucinate, occasionally run slowly or time out.

Google reports a 68.9% Vals Index score and 77.9% on DeepSWE v1.1. The same table shows a more mixed picture elsewhere: Argon trails several competitors on FrontierSWE v2, Terminal-Bench 4.0 and Terminal-Bench Science 0.1. These are vendor-published results spanning different harnesses and scoring methods, so they should not be read as proof of universal superiority.

Introductory API pricing is $2 per million input tokens and $10 per million output tokens, with cached input priced 95% lower. Google says standard pricing will rise to $4 and $20 respectively after November 30, 2026.

Sources

benchmarksGoogle DeepMindfrontier modelsGemini 4 Argonmultimodal AI