Gemini 4 Argon

Google DeepMind has unveiled its latest frontier model, Gemini 4 Argon, which is currently being piloted by select cybersecurity professionals via the company's Fairwind Program. Designed to excel in complex, multi-step workflows, Argon sets a new standard in reasoning and software engineering tasks. The model is accessible with an introductory pricing structure of $2 per million input tokens and $10 per million output tokens, offering a 95% discount on cached input tokens.

Gemini 4 Argon is already being utilized internally by thousands of Google employees, particularly for specialized coding tasks, extensive research, and high-quality document generation. To accommodate the model's extended output capabilities, Google has significantly increased the output token limit to an industry-leading 1 million tokens, up from the previous 64K tokens.

This enhanced capacity allows Argon to deeply analyze and generate content across long, intricate tasks. The model's performance has been thoroughly evaluated across various domains, including software engineering, enterprise knowledge work, and cybersecurity defense. In software engineering evaluations, Argon outperforms other models, particularly on DeepSWE v1.1, achieving a score of 77.9%.

Argon also leads in economic impact assessments, scoring highly on the Vals Index for finance, coding, legal, and tax work. Additionally, the model demonstrates superior performance in visual understanding tasks, including long video analysis, with a score of 91.7% on LVBench. Gemini 4 Argon has been specifically trained to excel in cybersecurity defense, with the ability to autonomously identify, validate, and patch critical software vulnerabilities.

The model is being made available to trusted cybersecurity professionals without the usual guardrails, enabling them to leverage its full frontier-level capabilities to tackle complex cyber threats.