Google announced Gemini 4 Argon on September 30, 2026, as its first frontier large language model since Gemini 3.1 Pro, with an immediate phased rollout limited to vetted cybersecurity defenders in the Fairwind Program. The model emphasizes long-horizon reasoning, achieving a leading 77.9% on DeepSWE v1.1 software engineering benchmarks and strong results on Vals Index enterprise tasks, while supporting an industry-leading 1 million output token limit. Google cited safety needs for the gradual expansion, including U.S. government pre-release reviews, before broader paid API and Google AI Ultra access. This limited launch, competitive positioning against GPT-6 Astra and Claude Opus 5.5, and absence of a firm public timeline create the core uncertainty for release-date or availability markets.