Google launches Gemini 4 Argon, its most advanced AI Model
Google has unveiled Gemini 4 Argon, the first model in its new Gemini 4 generation, as the company looks to close the gap with leading AI models from OpenAI and Anthropic. Google describes Argon as its new frontier model for complex, long-horizon workflows spanning software engineering, enterprise knowledge work and cybersecurity.
Unlike a conventional chatbot designed primarily for short interactions, Argon is built to sustain reasoning across much longer tasks. Google has expanded its output limit to 1 million tokens, up from 64,000 in the previous generation, giving the model substantially more room to work through complex problems in a single trajectory.
Built for Coding and Enterprise Work
Google says Argon is already being used internally by thousands of employees for debugging, research and large-scale software engineering projects.
In one example, Argon agents helped Google migrate C and C++ codebases to Rust, including work involving more than 800,000 lines of code in the Fuchsia Zircon kernel. Google also says the model helped optimize memory usage across its data centers, freeing more than 300 tebibytes of memory once the changes were deployed.
The model is also designed for professional knowledge work, including financial research, legal drafting and analysis of complex documents, charts and long videos.
Google says Argon achieved a 77.9% score on the DeepSWE v1.1 benchmark for real-world software engineering tasks. The company also reports leading performance on several finance, legal and video-understanding evaluations.
Autonomous Cybersecurity Defense and Vulnerability Remediation
Cybersecurity is perhaps Argon's most significant differentiator. Google trained the model specifically for defensive security work and says it can autonomously identify, validate and patch critical software vulnerabilities.
On the CWE-bench v1 vulnerability remediation benchmark, Google says Argon tied for first place with a score of 68%. The company has also highlighted the model's ability to resist indirect prompt-injection attacks.
Google is initially providing Argon to trusted cybersecurity defenders through its Fairwind Program, rather than releasing it broadly. The company says the phased approach is intended to strengthen safeguards and gather feedback before wider availability.
Wider Release Still to Come
Google has not announced a firm date for general public availability. When released commercially, Argon will start at $2 per million input tokens and $10 per million output tokens, with cached inputs receiving a 95% discount.
