Google Unveils Gemini 4 Argon: First Look for Cyber Defenders
Google (see details) has unveiled Gemini 4 Argon, its latest generation AI model since Gemini 3 last November. Unlike previous versions, Argon is being rolled out first to a small group of cybersecurity defenders through Google's Fairwind Program and the US government's pre-release access process.
Koray Kavukcuoglu, head of Google DeepMind, calls Argon the company’s “next era of frontier intelligence.” It's designed for complex workflows in software engineering, law, finance, and cybersecurity defense.
Key Features:
- Enhanced Performance: Argon achieves a new high of 77.9% on DeepSWE v1.1, a test of long software engineering tasks, and ties for first place (68%) on CWE-bench, a security flaw testing benchmark. It outperforms OpenAI's GPT-6 Astra in several coding and knowledge work benchmarks.
- Increased Token Limit: The model now allows up to 1 million input tokens and generates up to 1 million output tokens.
- Wider Availability: Following initial testing, Google will gradually expand access to Argon, ultimately making it available at an introductory price of $2 per million input tokens and $10 per million output tokens.
Doubts within Google:
While Google promotes Argon's capabilities, some internal employees remain skeptical. While the model excels on benchmarks, its performance in real-world applications, particularly coding tasks, is reportedly less impressive.
Competitive Landscape:
Argon's launch follows OpenAI's decision to cancel its next model due to safety concerns and a lawsuit over agents that hacked Hugging Face.