Google announces Gemini 4 Argon for complex long-running tasks, starting with limited access

Google announced Gemini 4 Argon on September 30, 2026, introducing a new frontier model designed for complex, long-running workflows in real-world software engineering, enterprise knowledge work such as legal and finance, and cybersecurity defense.

Gemini 4 Argon is not broadly available to consumers yet. Google is starting with a phased rollout through its Fairwind Program, where access is being provided first to a limited set of trusted cyber defenders.

What Gemini 4 Argon is designed to do

Google DeepMind positions Argon as a frontier model built to sustain deeper reasoning over complex, long-horizon tasks. The company highlights coding, enterprise workflows, multimodal understanding, computer use and defensive cybersecurity as major areas of focus.

Google has published benchmark results covering science and math, long-context tasks, computer use, multimodal understanding and cybersecurity. Those figures are vendor-reported results, so they should not be treated as independent verification of overall superiority over competing models.

U.S. availability is still limited

At launch, Gemini 4 Argon is being made available through the Fairwind Program to selected cybersecurity partners rather than to the general public. Google also says it is participating in the U.S. government’s voluntary pre-release model access process while it evaluates the model and expands safeguards.

Google says access will expand to developers, enterprises and consumers over time, but it has not announced a broad U.S. consumer release date as of October 1, 2026. In other words, most U.S. Gemini users cannot simply select Argon in the standard Gemini app today.

Introductory API pricing

Google says Gemini 4 Argon will launch at an introductory price of $2 per million input tokens and $10 per million output tokens. Cached input is expected to be priced at a 95% discount relative to standard input pricing.

That pricing is tied to the model’s broader rollout. Because general developer access has not yet begun, developers should verify final availability, regional access and production pricing when Google opens Argon more widely.

A rollout shaped by safety

Google is emphasizing safety controls alongside capability. Its model page describes protections against harmful requests, improved resilience to indirect prompt injection, monitoring intended to catch misaligned behavior and hardened sandbox environments for high-risk training and evaluation.

The cybersecurity focus makes the staged rollout especially important. A model that can help defenders find and fix vulnerabilities can also create new misuse risks if released without adequate controls, so Google is limiting access while gathering feedback from early testers.

KX3 perspective: this is about sustained work, not just better chat

The most important part of Gemini 4 Argon is not simply another benchmark increase. The broader shift is toward AI systems that can keep working across longer, multi-step tasks without losing the thread.

That matters for large codebases, research across many documents, enterprise operations and security analysis. Argon looks less like a conventional chatbot upgrade and more like infrastructure for increasingly autonomous agents. The real test, however, will come after developers and consumers gain broad access and can compare the model independently.

Announcement and sources

Announcement date: September 30, 2026

Primary sources: Google DeepMind — Gemini 4 Argon model page / Google — “Gemini 4 Argon: our next era of frontier intelligence”

Image: Unsplash / Brecht Corbeel

日本語版を読む