Google Begins Limited Rollout of Gemini 4 Argon

Google has begun rolling out Gemini 4 Argon, its latest frontier artificial intelligence model, to trusted cybersecurity defenders and internal teams while it continues strengthening safeguards before broader access. The initial outside users are members of Google’s Fairwind Program, which the company says has enrolled more than 650 organizations. Google has not announced a date for expanding access to developers, businesses or consumers. Google says Argon is designed for long, complex tasks in software engineering, business work and cybersecurity. The company reported scores of 77.9 on DeepSWE v1.1, a benchmark for extended software engineering tasks, and 51.3 on AutomationBench, which tests end-to-end business work. Google also raised the model’s output limit to one million tokens, from 64,000 in earlier Gemini models. Those figures come from Google’s own evaluations. In a separate assessment, Artificial Analysis gave Argon a score of 53 on its Intelligence Index at the model’s highest available reasoning setting, tying it with OpenAI’s GPT-6 Astra at its maximum setting. Argon’s overall score was 42, compared with Astra’s 43.