The Cyber Hunt: Gemini 3.8 Flash and its Agent Twin

Google has recently unveiled two new versions of its 3.8 Flash model, continuing its trend of rapid innovation in the AI space. The new variants include a standard Flash model designed for agentic tasks, software development, and multi-step reasoning, as well as Flash Cyber, optimized for vulnerability detection and mitigation.

According to Google CEO Sundar Pichai, the 3.8 Flash model represents a significant improvement over its predecessor, particularly in areas such as software engineering, agentic tasks, and multi-step reasoning. The Flash Cyber model, on the other hand, is being touted as Google’s most capable cybersecurity model, excelling in discovering vulnerabilities and patching them at scale.

The 3.8 Flash release marks Google’s third Flash model in just six weeks, following closely on the heels of version 3.7. The new model is now available in Gemini Enterprise, offering developers the opportunity to explore its capabilities through various Google platforms.

Priced at $0.75 per million input tokens and $3.75 per million output tokens, the 3.8 Flash model allows users to customize and adjust the level of effort based on their specific needs. This flexibility enables users to balance factors such as quality, cost, and latency according to their requirements.

One key feature of the 3.8 Flash model is its ability to handle complex tasks with greater diligence, such as executing additional reasoning steps. While the model may use more tokens for optimal performance in some cases, it offers a range of capabilities, including text, image, audio, video, and PDF file processing.

The 3.8 Flash model has been extensively evaluated across various benchmarks, showcasing its prowess in coding, multimodal capabilities, computer usage, long-context reasoning, and scientific reasoning. It has demonstrated superior performance in specialized knowledge domains, outperforming its predecessor and other frontier models in finance and legal benchmarks.

In practical applications, the 3.8 Flash model has been used to create interactive games, visualizations, and maps, highlighting its versatility and creativity. Its performance has been recognized in AI rankings, placing it ahead of competitors in areas such as text processing and multi-step reasoning tasks.

Meanwhile, the Flash Cyber model is being rolled out to select partners through Google’s Fairwind Program, offering advanced cybersecurity defense capabilities. The model has undergone rigorous training in cybersecurity and is designed to excel in vulnerability discovery and automated patching, prioritizing defense over offensive capabilities.

Google is already leveraging the capabilities of the 3.8 Flash Cyber model to secure its own code, demonstrating its effectiveness in identifying and patching vulnerabilities. The model has shown promising results in vulnerability detection, outperforming larger commercial models in terms of accuracy and cost efficiency.

Overall, Google’s latest Flash models represent a significant leap forward in AI technology, offering advanced capabilities in software development, cybersecurity, and beyond. The company continues to push the boundaries of innovation, delivering cutting-edge solutions to meet the evolving needs of the digital landscape.