Google announced Gemini 4 Argon on September 30, 2026, positioning it as a frontier large language model optimized for long-horizon software engineering, enterprise workflows in legal and finance, and cybersecurity defense, with an expanded 1-million-token output limit. The model rolled out first under the Fairwind Program to vetted cyber defenders for safety testing, reflecting Google’s phased release strategy amid ongoing U.S. government pre-release reviews. It leads several benchmarks like DeepSWE v1.1 at 77.9% and the Vals Index for economic impact tasks, outperforming recent rivals such as OpenAI’s GPT-6 Astra and Anthropic’s Claude Opus 5.5 on Google’s reported metrics, though it trails on select coding tests. Broader API access for paid users and Google AI Ultra subscribers remains the next key catalyst, with no firm public timeline yet after the cancellation of the prior Gemini 3.5 Pro effort.