Google Gemini 4 Argon released: 1 million output tokens and a focus on cybersecurity, coding, and agentic capabilities

The most notable technical feature of the new Gemini 4 Argon AI model is its 1-million-token output window, up from 64,000 previously. This is designed to allow Argon to plan, analyze, and operate autonomously for longer periods, particularly when handling complex and time-consuming tasks. With a model like this, there would be no need to break a complex task down into a series of individual prompts.

With this, Google is taking another step toward autonomous agents. The agent is given a task, plans its approach over an extended period, analyzes code and information, and then carries out various intermediate steps, iteratively testing and correcting their results before ultimately completing the task.

Google emphasizes that the model is also used extensively in-house. At the company, Argon agents are helping quantum computing researchers optimize Spacetime resources, improving memory optimization in data centers, porting C/C++ code to Rust, and optimizing video decoders written in Rust for speed.

At the time of publication, Argon is not yet publicly available. It will initially be made available to trusted testers and cybersecurity experts. Access is then expected to gradually expand to developers, businesses, and consumers, initially for paying API customers and Google AI Ultra subscribers. The announced introductory price is $2 per 1 million input tokens and $10 per 1 million output tokens. After the introductory period, those prices will rise to $4 and $20, respectively; cached input tokens are 95% cheaper.