Introducing Gemini 3.8 Flash and Flash Cyber

Google has announced Gemini 3.8, introducing two new model variants designed for advanced reasoning, coding, and cybersecurity tasks. Building on previous releases, the new models maintain low latency and pricing while delivering significant performance improvements across specialized domains.

Gemini 3.8 Flash for agents and coding

Gemini 3.8 Flash is designed as a high-performance workhorse model for software engineering and multi-step reasoning. It builds on earlier versions by executing extra reasoning steps and using tools iteratively on complex tasks.

On the HLE-Verified benchmark, the model achieves a 54.9% score, demonstrating its capability across STEM and humanities fields. It is available to developers at $0.75 per million input tokens and $3.75 per million output tokens.

Gemini 3.8 Flash Cyber for security

The second variant, Gemini 3.8 Flash Cyber, is built specifically for vulnerability detection and automated patching. The model is trained to assist defenders rather than perform offensive operations.

On the CWE-Bench patching benchmark, Flash Cyber achieves a pass@1 score of 47.2%. Early deployment inside Google and with external partners shows it producing correct patches for Chrome vulnerabilities and identifying issues across complex codebases spanning 20 programming languages.

Access and availability

Gemini 3.8 Flash is available immediately through the Gemini API, Google AI Studio, Android Studio, and enterprise platforms. The specialized Flash Cyber variant is restricted to trusted defenders, government authorities, and critical infrastructure operators through a dedicated access program.