Google has introduced Gemini 4 Argon, a new frontier model designed for coding, office tasks, and cyber defense. The model achieved a 77.9% score on the DeepSWE v1.1 benchmark, which evaluates software engineering capabilities. It supports a context window of up to 1 million tokens, a significant increase from the 64,000 tokens supported by previous versions.
In cybersecurity testing, Gemini 4 Argon recorded a 0.7% success rate on the Gray Swan Indirect Prompt Injection benchmark. This performance surpasses Claude Opus 5.5 and Claude Fable 5.1, which scored 1.0%, as well as GPT-6 Astra at 8.5%. The model also achieved a 70.9% success rate on the internal Wiz Penetration Test Benchmark, improving upon the 58.2% result of the previous Gemini 3.8 Flash Cyber model.
The model is currently available to vetted security teams through the Fairwind Program, an initiative launched on September 2 that includes over 650 partners. Google stated that the model is provided without standard cyber guardrails to assist in identifying vulnerabilities. The company is also participating in the U.S. government's voluntary pre-release model access process.
Google has set introductory pricing for Gemini 4 Argon at $2 per million input tokens and $10 per million output tokens. Standard rates are set at $4 and $20 per million tokens, respectively. The release follows the company's previous launch of smaller Flash models in July and coincides with Google's participation in a voluntary AI accord.
