Google Releases Gemini 3.8 Flash and Cyber Versions

Google has released the Gemini 3.8 series, which includes the general-purpose model Gemini 3.8 Flash and the cybersecurity-focused model Gemini 3.8 Flash Cyber. Both models share identical foundational capabilities and leverage long-running agent loops to repeatedly evaluate and improve performance. Targeted at long-cycle software engineering, autonomous agents, and multi-step reasoning in specialized domains, Google states that Gemini 3.8 Flash shows significant improvements over 3.7 Flash on DeepSWE and financial/legal agent benchmarks, achieving an HLE-Verified score of 54.9%. The initial pricing remains unchanged at $0.75 per million input tokens and $3.75 per million output tokens; however, complex tasks may invoke additional tools and consume more tokens, while developers can reduce inference intensity to manage costs. Designed primarily for vulnerability discovery and automated patching, Gemini 3.8 Flash Cyber achieved success rates exceeding 70% in internal vulnerability tests covering 20 programming languages. It scored 47.2% pass@1 on CWE-Bench, nearing leading models’ 47.8%, and generated correct security patches for Chrome at 2.6 times the volume produced by larger commercial models. Due to broader safety restrictions, this release is exclusively accessible via the Fairwind Program to approved government agencies, critical infrastructure operators, and software maintainers. The general-purpose variant is now live across platforms including the Gemini API, Google AI Studio, Android Studio, and Gemini Enterprise, and is available to users subscribed to Google AI Pro and Ultra plans.