Google Launches Gemini 4 Argon AI Model on September 30, 2026

What happened

On September 30, 2026, Google officially announced its new frontier artificial intelligence model, Gemini 4 Argon ^[1][5][6][7]. Presented by Google DeepMind SVP and Chief AI Architect Koray Kavukcuoglu, Argon is engineered for deep reasoning across complex, long-horizon workflows spanning real-world software engineering, enterprise knowledge work in legal and finance, and cybersecurity defense ^[1][5].

To support extensive reasoning and long-form analysis, Google expanded the model's output token limit to an industry-leading 1 million tokens, a significant increase from the previous 64K limit ^[1][6]. In evaluations, Argon established strong performance benchmarks, scoring 77.9% on DeepSWE v1.1 and ranking high on various domain-specific indexes including Vals Finance Agent v2 and Harvey's Legal Agent Benchmark ^[1][6].

Safely releasing these capabilities prompted a phased rollout. Google initially made Argon available to a selected group of trusted cyber defenders through its Fairwind Program without cyber guardrails, while engaging with the U.S. government’s voluntary pre-release model access process ^[1][5][6][7]. For broader availability, the model launched with an introductory pricing structure of $2 per million input tokens and $10 per million output tokens, with cached input tokens discounted by 95% ^[1][6].

Why it matters

Gemini 4 Argon introduces advanced capabilities designed to alter internal and external engineering productivity. Inside Google, thousands of employees have utilized the model for specialized coding tasks, research, and memory efficiency improvements, such as freeing over 300 TiB of memory across data centers ^[1][6]. Argon agents have also undertaken large-scale codebase migrations, translating C/C++ codebases to Rust across critical systems including the Fuchsia OS Zircon kernel and the libgav1 video decoder ^[1][6].

In defensive cybersecurity, Argon was trained to autonomously find, validate, and patch critical software vulnerabilities ^[1][7]. Programs like Wiz's Scan for Good initiative have deployed Argon to uncover severe risks in infrastructure, including critical exposures in healthcare software ^[1]. To protect against potential risks, Google implemented strict safety measures prior to broad release, incorporating defenses against misuse, indirect prompt injection attacks, and continuous monitoring of internal activations and chain-of-thought execution to guard against misalignment ^[1][6].

What to watch

As Google transitions Gemini 4 Argon from its initial restricted deployment to broader availability for Google AI Ultra subscribers, enterprise customers, and paid API users, observers will monitor the expansion of its safeguards ^[1][6]. Key areas to track include how the model performs under commercial developer loads, the evolution of the Fairwind Program's cybersecurity deployments, and how competing AI labs respond to Argon's benchmark scores across software engineering and enterprise automation ^[1][5][6][7].

Sources

1. Introducing Gemini 4 Argon blog.google
2. Google announces Gemini 4 and says it’s so capable that only ‘trusted cyber defenders’ can have it right now | The Verge theverge.com
3. Google announces Gemini 4 Argon as its new frontier model 9to5google.com
4. Google releases Gemini 4 Argon, called its most powerful model yet | TechCrunch techcrunch.com