Google Unveils Gemini 4 Argon AI Model With Advanced Coding and Cybersecurity Features

Google has launched Gemini 4 Argon, a new AI model it says is a big step up in coding, professional tasks, and cybersecurity.

Key Highlights

• Stronger coding and cybersecurity skills
• Scored 68.9 percent on the Vals benchmark and 91.9 percent on Vibe Code Bench
• Access limited for now to trusted cyber defenders and the US government through the Fairwind program
• Introductory pricing starts at $2 per million input tokens and $10 per million output tokens

What It Can Do

Argon is built for deep reasoning across complex workflows and supports an output limit of 1 million tokens. Google says it matches or beats rivals like GPT-6 Astra and Claude Fable 5.1 in several areas. On the Vals benchmark it scored 68.9 percent, ahead of Fable 5.1 (65.8) and GPT-6 Astra (63.1). On Vibe Code Bench it hit 91.9 percent, against 90.3 and 89.6. It did not give good result in every test  including Terminal-bench 4.0.

Cybersecurity and Early Access

Google says Argon can find, confirm, and patch critical software flaws on its own. Wiz, Google's security arm, is already using it, and in early demos it caught a serious vulnerability in hospital software that earlier models missed. For now the access is restricted, similar to OpenAI's Daybreak and Anthropic's Project Glasswing. Google CEO Sundar Pichai says the model ships with frontier safeguards and its access will expand after the tester's feedback.

Pricing and Availability

Paid API customers and Google AI Ultra subscribers come next. Its launch pricing is $2 (about Rs 190) per million input tokens and $10 (about Rs 950) per million output tokens, with a 95 percent discount on cached input. Afterward, prices rise to $4 (about Rs 380) and $20 (about Rs 1,900). There is no public release date yet.