Google unveiled Gemini 4 Argon, a frontier AI model aimed at complex, long-duration workflows in software engineering, enterprise research, and cyber defense, with a phased rollout to vetted security partners through its Fairwind Program. The company set introductory pricing at $2 per million input tokens and $10 per million output tokens, with a 95% discount on cached inputs, and expanded the model’s output limit to 1 million tokens to support deeper reasoning. Internally, Argon has driven gains from large codebase migrations to Rust—including work on the Fuchsia Zircon kernel—to quantum algorithm optimization and sizable data-center memory savings. On benchmarks, Argon reports state-of-the-art or leading scores on DeepSWE v1.1, the Vals Index, LVBench, and CWE-bench v1, and Google says it will release uncapped cyber capabilities to trusted defenders. The company is coordinating with U.S. government voluntary pre-release access programs and tightening safeguards against misuse, prompt injection, and misalignment, while hardening sandbox environments. Broader availability is planned for paid API customers and Google AI Ultra subscribers following further testing.
Related articles:
NIST AI Risk Management Framework (AI RMF 1.0)
GPT-4 Technical Report
