Google released Gemini 3.8 Flash, a reasoning and coding model aimed at software engineering and autonomous agent tasks, and Gemini 3.8 Flash Cyber, a cybersecurity-specialized variant, according to Google's own announcement.
Flash Cyber shows a success rate above 70% on Google's internal vulnerability discovery benchmarks across 20 programming languages and produces 2.6 times more correct patches than leading commercial models on the CWE-Bench patching benchmark, Chrome's security team found, Google said. Google is restricting the Cyber variant to vetted governments and partners through its Fairwind Program rather than releasing it broadly.
The general 3.8 Flash model keeps the same introductory pricing as 3.7 Flash, $0.75 per million input tokens and $3.75 per million output tokens, rising to $1.50 and $7.50 after December 31, Google said. Google scored the model at 54.9% on HLE-Verified, a multi-step reasoning benchmark spanning STEM, humanities and professional domains, and said it outperforms larger frontier models on coding and financial and legal analysis benchmarks.
The model is available now in Google AI Studio, Android Studio, Gemini Enterprise, the Gemini app, Search's AI Mode and Google Sheets, according to Google.
Gating the sharper cybersecurity capability while shipping the general model to everyone is the same pattern OpenAI used with Astra this week. Two labs racing on frontier capability are independently landing on the same answer: ship the productivity model broadly, keep the offense-capable one on an approval list.