Google DeepMind Launches Gemini 3.8 Flash and Restricted Cyber Variant for Automated Security

Google DeepMind has introduced Gemini 3.8 Flash and Gemini 3.8 Flash Cyber, marking the third Flash release in just six weeks and arriving three weeks after the release of Gemini 3.7 Flash. [1][4]

Both new variants are powered by the same foundational intelligence, which was refined through long-running agentic loops that recursively evaluate the underlying models. [1][4] The standard Gemini 3.8 Flash model is positioned for long-horizon software engineering and agentic tasks. [4] It is available at an introductory price of $0.75 per million input tokens and $3.75 per million output tokens, which runs until December 31. [3]

The standard Gemini 3.8 Flash model is generally available today for production traffic through the Gemini API, Google AI Studio, Android Studio, Antigravity, and Gemini Enterprise. [1] The model specifications feature a 1,048,576-token context window alongside a maximum output of 65,536 tokens. [1] The model weights are closed, meaning there is no on-premises or self-hosted deployment path available. [1]

The model offers thinking levels of LOW, MEDIUM, and HIGH, with MEDIUM serving as the default setting. [1] Developers migrating to the new version should note that the MINIMAL thinking level is no longer supported on 3.8 Flash, and attempting to set it will return an API validation error. [1]

Unlike the generally available standard model, Gemini 3.8 Flash Cyber is not openly deployable and is restricted to trusted defenders. [1][4] Access to the cybersecurity-focused variant is granted on a case-by-case basis through Google's new Fairwind Program. [1][4] The Cyber model is designed to deliver frontier-level performance in vulnerability detection and automated patching. [4]

Google states that the standard Gemini 3.8 Flash model beats competitors such as Claude Opus 5 and GPT-5.6 Sol on some benchmarks. [2] Specifically on the DeepSWE v1.1 benchmark for long-horizon software engineering, the 3.8 Flash variant outperforms most larger frontier models. [4]

Google's aggressive release cadence—three Flash models in six weeks—indicates a rapid iteration cycle focused on reasoning and coding capabilities rather than architectural overhauls. By maintaining the same price point as version 3.7 while claiming performance that beats rival models like Claude Opus 5 and GPT-5.6 Sol on some benchmarks, Google is competing fiercely on cost-to-performance ratios for agentic workflows. The bifurcation into a generally available model and a restricted Cyber variant highlights the growing security sensitivities around automated vulnerability detection and patching tools. The removal of the MINIMAL thinking level suggests a baseline compute floor Google requires to ensure quality outputs. What the sources don't address: how exactly Google defines a "trusted defender" for inclusion in the restrictive Fairwind Program.