Google Unveils Gemini 4 Argon AI Model With 1 Million-Token Output Limit

Google announced its new flagship AI model, Gemini 4 Argon, on September 30, 2026, after months of delays. The model anchors Google’s Gemini 4 generation and is positioned as the company’s most advanced yet, designed for complex, long-horizon workloads.

Google describes Gemini 4 Argon as delivering frontier-level performance in real-world software engineering, enterprise knowledge work, including legal and finance tasks, and cybersecurity defense.

Gemini 4 Argon size and capabilities

Argon is larger than Google’s previous top-tier “Pro” models. It features a significantly expanded output token limit of 1 million tokens, up from 64K, to support deeper reasoning on extended tasks.

Google claims strong results on internal and third-party benchmarks, including leading or competitive scores against OpenAI’s GPT-6 Astra and Anthropic’s latest models on coding, cybersecurity, including CWE-bench, knowledge work through the Vals Index, and other evaluations.

Limited initial access

Due to its capabilities, particularly in cybersecurity, Google is initially rolling out Gemini 4 Argon only to a select group of trusted cybersecurity partners through its Fairwind Program.

The company is also participating in the U.S. government’s voluntary pre-release access process.

Broader availability to developers, enterprises, Google AI Ultra subscribers and the public is planned later, following additional testing and guardrail refinements. No exact public release date was given.

Gemini 4 Argon pricing

When the model becomes more widely available, introductory rates are set at $2 per million input tokens and $10 per million output tokens, with steep discounts on cached inputs.

Delays before Gemini 4 Argon

Google had not released a major new flagship model since the Gemini 3 series in late 2025 and early 2026. A planned Gemini 3.5 Pro update, promised by CEO Sundar Pichai for around June 2026, was never released after internal performance issues.

In the interim, the company focused on smaller and faster Flash models while rivals OpenAI and Anthropic advanced with new frontier releases.

Leadership changes at Google DeepMind, including Demis Hassabis stepping aside, also occurred during this period.

Gemini 4 Argon is Google’s bid to close the gap. The model is already in use internally at Google for coding, research and other complex tasks. Official details are available on the Google blog and DeepMind site.