Google DeepMind released Gemini 3.8 Flash on Tuesday — its third Flash-tier model in just six weeks — alongside Gemini 3.8 Flash Cyber, a security-hardened variant that is available exclusively to trusted defenders through a new program called Fairwind.
Gemini 3.8 Flash is positioned as Google's most capable Flash-tier model to date, designed for agentic workflows and multi-step reasoning tasks. The standard model is priced at $0.75 per million input tokens and $3.75 per million output tokens through December 31, 2026, an introductory rate that undercuts many competitors.
The more notable release is Gemini 3.8 Flash Cyber, which Google says achieves 70%+ vulnerability detection accuracy across 20 programming languages. In internal testing against Chrome vulnerabilities, it delivered 2.6x more correct patches than leading commercial models, and posted 47.2% pass@1 on CWE-Bench, a standard cybersecurity evaluation benchmark.
Flash Cyber is available only through the Fairwind Program, Google's initiative to provide frontier AI security tools to trusted defenders — a model reminiscent of how some governments restrict access to offensive cyber tools. Google did not disclose pricing for the Cyber variant.
The rapid cadence of Flash releases — 3.7 shipped August 13, 3.8 arrived September 2 — reflects Google's strategy of maintaining a roughly one-model-per-month release schedule under CEO Sundar Pichai. WSJ separately reported that Gemini 4 has performed well in pre-training evaluations but still requires post-training work.
The cybersecurity-only model represents a growing trend of AI companies creating restricted-access variants of their models for sensitive applications, balancing openness with safety concerns in an era of increasingly capable AI systems.
