Google has officially launched its most powerful model, Gemini 4 Argon, making it available to developers, enterprise customers, and subscribers worldwide. The model was designed to solve the most intractable real-world problems in coding, finance, law, cybersecurity, and the frontiers of science, delivering substantially greater accuracy and depth than the previous generation on multi-step complex tasks. It has already claimed the top spot in Vals Index, an authoritative benchmark measuring key economic areas such as finance, code, law, and tax.
In terms of underlying specifications, Gemini 4 Argon has received an epic upgrade, for the first time supporting long-decoding output of up to 1,048,576 Tokens (about 1 million tokens). Note that its input, reasoning/thinking, and final output share the same 1 million Token context window, enabling it to directly generate extremely large bodies of long-form text, complex codebases, or in-depth research reports.
Real-world testing shows that the model has already demonstrated formidable productivity in engineering and frontier research. In software engineering, Google's Chrome team used Gemini 4 Argon to carry out an architectural refactoring of tens of millions of lines of code, compressing work that would originally have taken months into weeks; in addition, the model autonomously completed performance optimization of an open-source project overnight, successfully reducing memory session preparation time by 40% and improving cold import speed by 21%.
To promote applications in frontier research, Google also announced it will provide 10,000 scientists with free access to the top tier of Google Antigravity for 6 months. In cybersecurity, Google has further expanded the scope of the Fairwind program, helping security professionals use the model to efficiently find and fix critical vulnerabilities.
On the safety and protection front, Google has deployed an entirely new multi-layered defense system for the model, achieving significant improvements in resistance to jailbreak attacks, prompt injection protection, and agent alignment monitoring, with the violation rate for harmful requests nearly cut in half compared with the previous Gemini models.
In terms of commercialization and pricing, Google has launched a limited-time introductory promotional price (until November 30, 2026): input at $2 per million Tokens, output at $10 per million Tokens, and cached input at $0.1 per million Tokens. After the promotional period ends, regular pricing will be adjusted to: input at $4 per million Tokens, output at $20 per million Tokens, and cached input at $0.2 per million Tokens.
Currently, developers and enterprise customers can access the model through Google AI Studio and the Gemini API, and subscribers to Google AI Ultra and Pro can also experience this latest model in the Gemini app, Search AI Mode, and the Google Antigravity platform.