Google has introduced Gemini 4 Argon, an AI model designed for complex workflows, software engineering, and cybersecurity. The system supports an output of up to 1 million tokens, a significant increase from the 64,000 tokens supported by previous versions.
The model is currently available only to a select group of cyber defenders via the Fairwind Program. Google plans to expand access following safety evaluations and consultations with the U.S. government. The company has set introductory API pricing at $2 per million input tokens and $10 per million output tokens, with future plans to increase these rates to $4 and $20 respectively.
Argon demonstrated high performance on several benchmarks, achieving 68% on CWE-bench v1 for security flaw remediation and 77.9% on DeepSWE v1.1 for software engineering tasks. It also ranked first on the Vals Index and scored 51.3% on Zapier’s AutomationBench.
Internal testing shows the model can optimize data-center memory, potentially saving between 500 TiB and 1 PiB of capacity. Additionally, researchers used Argon to improve quantum computing algorithm baselines by 40% within minutes and to assist in migrating C and C++ codebases to Rust.
