Google announced Gemini 4 Argon on September 30, 2026, as its new frontier large language model optimized for long-horizon software engineering, enterprise tasks in legal and finance, and cybersecurity defense. The model features a 1-million-token output limit and leads benchmarks such as DeepSWE v1.1 at 77.9% while tying or topping rivals including Claude Opus 5.5 and GPT-6 Astra on several coding and vulnerability remediation tests. Initial access is limited to vetted partners in the Fairwind Program and internal Google teams amid ongoing U.S. government safety reviews, with broader API availability and introductory pricing of $2/$10 per million input/output tokens planned afterward. This phased rollout and competitive positioning against OpenAI and Anthropic models are shaping trader views on release timing and adoption milestones.