Google Launches Gemini 4 Argon for Complex AI Workloads

Google has unveiled Gemini 4 Argon, a new frontier artificial intelligence model designed to sustain deep reasoning across complex, long-horizon workflows. The company is initiating a phased rollout through its Fairwind Program, beginning with trusted cyber defenders and U.S. government partners engaged in a voluntary pre-release access process. This controlled deployment allows Google to gather real-world feedback and refine safety guardrails before expanding availability to developers, enterprises, and consumers. Engineered to handle high-stakes tasks, Argon delivers frontier performance in software engineering, legal, financial, and cybersecurity operations. The model is already accelerating internal infrastructure projects, including large-scale migrations of C and C++ codebases to Rust. In one documented optimization of the open-source libgav1 video decoder, Argon agents replaced 32,000 lines of SIMD code with automatically vectorized, memory-safe Rust, achieving a 2.7x performance improvement while matching original output fidelity. The model has set new state-of-the-art benchmarks in real-world software engineering, surpassing previous records on DeepSWE v1.1, while also leading the Vals Index for economic impact across finance, legal, and tax work. Additionally, Argon ranks first on Zapier’s AutomationBench for end-to-end business execution and achieves top results on LVBench for long-form video understanding. To manage the risks associated with frontier capabilities, Google has implemented comprehensive safety protocols. The model is built to refuse requests related to cyber or chemical, biological, radiological, and nuclear threats, while preserving legitimate scientific research. Internal robustness checks utilize advanced monitoring of internal activations to detect potential misuse. Argon also demonstrates industry-leading resilience against indirect prompt injection attacks, validated through adversarial training and third-party red team assessments. Commercially, Google has structured Argon’s pricing at $2 per million input tokens and $10 per million output tokens, with a 95 percent discount applied to cached inputs. As the model advances through its iterative testing phase, Google plans to scale access to broader enterprise and consumer markets, positioning Argon as a foundational tool for next-generation automation and knowledge work.