Google Unveils Gemini 4 Argon: AI Powerhouse for Complex Tasks

Introduction

Google has launched Gemini 4 Argon, a new AI model designed to handle complex tasks across multiple domains, including coding, finance, law, cybersecurity, and science. The model is rolling out globally through the Gemini API, AI Studio, and Google's enterprise offerings.

Features and Benefits

API access launches at an introductory price of $2 per million input tokens and $10 per million output tokens. Cached inputs receive a 95% discount. Gemini 4 Argon supports a unified input and output limit of one million tokens, allowing it to carry out longer tasks involving large amounts of information and multiple tools.

Applications and Use Cases

The release expands Google's push into AI systems capable of sustained professional work. According to the company, Argon ranks first on the Vals Index, which measures performance across economic sectors including finance, coding, legal and tax work. Google pointed to internal engineering projects as examples of the model's capabilities. Over the past few weeks, Argon in Antigravity helped teams increase agentic code submissions by more than 35% while reducing code rollbacks by nearly a third.

Early Adopter Feedback

Early enterprise users reported improvements in their own evaluations. Box said Argon increased accuracy by 14% compared with Gemini 3.1 Pro on enterprise knowledge tasks. Financial analysis company Kepler said Gemini 4 matched more than 99% of figures in its analyst-built reference models.

Beyond Software and Business Workflows

Google highlighted applications in quantum computing, nuclear fusion, rare disease research and enzyme design. In an enzyme research project, Argon assembled data, tuned models and helped identify five candidate designs within hours. Google said subsequent laboratory testing confirmed catalytic activity across those designs.

Expanded Safeguards and Future Plans

Google also described a robotics demonstration in which Argon watched a video of a person playing cornhole, calculated the necessary movements and generated simulation code. When transferred to Apptronik's Apollo humanoid, the robot landed seven consecutive shots in a real-world trial, according to the company. Argon is the first release in the Gemini 4 family, with additional models planned.

Conclusion

The launch includes expanded safeguards against misuse, unauthorized agent actions and malicious instructions embedded in external content. Google said Argon recorded a 0.7% attack success rate on the Gray Swan Indirect Prompt Injection benchmark across 15 adaptive attempts.