Google launches Gemini 4 Argon to reclaim the AI frontier

On Wednesday, Google unveiled its latest AI model, Gemini 4 Argon, marking the beginning of the Gemini 4 era. This new model is positioned as a significant advancement for frontier intelligence, according to the tech giant.

Gemini 4 Argon excels in tasks such as coding, cybersecurity defense, and enterprise knowledge work, particularly in sectors like legal and finance. It aims to reassert Google's position in the competitive AI landscape, which has seen rivals like OpenAI and Anthropic dominate areas such as engineering and cybersecurity.

The company's previous attempt, Gemini 3.5 Pro, was postponed due to performance issues and ultimately not released. However, Google's new model, Argon, has already shown impressive results. It is being made available to a select group of trusted partners within Google's Fairwind Program, which includes vetted governments and cyber authorities for proactive testing and addressing cyber vulnerabilities.

Argon has matched OpenAI's GPT-6 Astra in the CWE-bench, a test measuring models' ability to find and patch security vulnerabilities. It also sets a new high score on a test assessing real-world, long-horizon engineering tasks. While the exact launch date for Google Gemini 4 Argon hasn't been announced, the company is actively engaged in the US government's early-access framework for assessing cybersecurity and other risks associated with frontier models before its release.

Inside Google, employees are already leveraging the new model for tasks like debugging and large-scale codebase migrations. Google claims that Argon has helped engineers save over 300 tebibytes of memory across its data centers without requiring new hardware. To ensure the model's safe operation, Google has implemented systems to monitor Argon's chain-of-thought and halt its functioning if necessary.

These safety measures also include precautions around interacting with the model to prevent it from learning to evade monitoring.