Google Launches Gemini 3.8 Flash with Cybersecurity Focus

Google has unveiled the latest version of its AI model, Gemini 3.8 Flash, the third release in the Flash series over the past six weeks. The update prioritizes enhanced intelligence for agent-based workflows and cybersecurity. The model is available in two variants: a standard version for developers and enterprise clients, and a specialized Gemini 3.8 Flash Cyber edition for information security tasks.

According to Google, Gemini 3.8 Flash outperformed larger models in several benchmarks. In the DeepSWE v1.1 test, which evaluates complex engineering problem-solving, the model achieved 73.7%, closely trailing Claude Opus 5 (74%) and significantly surpassing the previous Gemini 3.7 Flash. The model also demonstrated high efficiency in financial analysis, legal processes, and long-form video processing.

Gemini 3.8 Flash remains one of the most affordable options on the market, with input tokens priced at $0.75 per million and output tokens at $3.75. Future pricing adjustments will increase these rates to $1.50 and $7.50, respectively. The cybersecurity-focused version, Gemini 3.8 Flash Cyber, is exclusively available to participants in the Fairwind program, which targets government entities and critical infrastructure operators. This model achieved record-breaking scores of 86.2% in C/C++ vulnerability detection tests and 71% in real-world scenarios across 20 programming languages.

A key innovation in Gemini 3.8 Flash Cyber is its ability to automatically patch vulnerabilities. Google reports that the model generates 2.6 times more accurate patches for Chrome vulnerabilities than comparable commercial solutions. The standard version of Gemini 3.8 Flash is now accessible to developers, enterprise clients, and subscribers of Google AI Pro and Ultra.