In chemistry, argon is the noble gas that famously doesn't react with anything. Google has given that name to a model it hopes will cause a big reaction. It's a bold choice, and this time the benchmarks are doing a lot of the talking.
Built for the Hard, Multi-Step Work
Google launched Gemini 4 Argon on October 13, calling it its latest frontier model and pitching it at complex, multi-step work in coding, finance, legal, cybersecurity, and science. It's available through the Gemini API in AI Studio and Google Cloud, through Gemini for business and Google Antigravity for enterprises, and in the Gemini app and AI Mode in Search for AI Pro and Ultra subscribers.
Introductory pricing is $2 per million input tokens and $10 per million output tokens, with cached input 95% off. That price doubles to $4 and $20 on November 30, 2026. Google also claims the top spot on the Vals Index, which covers finance, coding, legal, and tax tasks, along with 87% on the ERQA embodied-reasoning benchmark. As always, those are Google's own numbers.
The Prompt-Injection Number Matters Most
A robot sinking seven straight cornhole shots makes a great demo. For businesses, though, the more important line is a quieter one: Google reports 0.7% attack success on the Gray Swan indirect prompt-injection benchmark. Google also says Argon's agentic features pause for human confirmation before irreversible steps. If you're connecting an AI agent to your CMS, CRM, or customer inbox, those are the numbers that should guide your decision.
The point many people will miss is the pricing schedule. A launch discount followed by a doubling is a reminder to build AI integrations that can switch models. Teams that hard-code one vendor's API throughout their stack will learn this the expensive way on December 1.
Argon is impressive. The smarter move is still building systems that can take advantage of whichever model is best next quarter.
Thinking about connecting a frontier model to your website or content workflows without locking yourself into one vendor's pricing? That's the kind of AI integration we design.
Source: Google Blog