Google DeepMind Launches Gemini 3.5 Flash Cyber, Finds 55 V8 Engine Vulnerabilities

Google DeepMind launched Gemini 3.5 Flash Cyber on July 21, 2026, a new cost-efficient AI security model specifically fine-tuned to find and patch code vulnerabilities, which has already identified 55 unique confirmed vulnerabilities in Chrome's V8 JavaScript Engine. For broader context, explore our AI News.

Gemini 3.5 Flash Cyber: A New Era in AI Cybersecurity

Google DeepMind's introduction of Gemini 3.5 Flash Cyber on July 21, 2026, signals a strategic expansion into AI-powered cybersecurity. This lightweight model is specifically fine-tuned for the intricate task of discovering and rectifying code vulnerabilities. Its design prioritizes efficiency in security applications, aiming to provide a focused solution for complex challenges in software integrity.

Access to Gemini 3.5 Flash Cyber will be restricted to a limited pilot program, exclusively available to governments and trusted partners. This controlled deployment will occur through Google's CodeMender security agent, indicating a deliberate approach to integrating advanced AI into sensitive security infrastructures.

Performance and Impact: V8 Engine Success

The effectiveness of Gemini 3.5 Flash Cyber has been demonstrated through its performance in real-world scenarios. The model successfully identified 55 unique confirmed vulnerabilities within Chrome's V8 JavaScript Engine. This achievement highlights its practical utility in detecting critical security flaws in widely used software components.

Furthermore, when invoked via CodeMender, Gemini 3.5 Flash Cyber exhibited competitive performance against larger models on the CyberGym benchmark. This suggests that its specialized tuning allows it to achieve significant results in vulnerability detection, even as a more lightweight solution.

Broader Gemini Updates: 3.6 Flash and 3.5 Flash-Lite

Alongside the cybersecurity model, Google also released Gemini 3.6 Flash and Gemini 3.5 Flash-Lite. These models expand the Gemini family, offering different capabilities and pricing structures to cater to various computational needs.

• Gemini 3.6 Flash: This model is priced at $1.50 per million input tokens and $7.50 per million output tokens. It offers improved capabilities across coding, knowledge work, and multimodal performance, while also reducing token usage by up to 17% compared to its predecessor, Gemini 3.5 Flash.
• Gemini 3.5 Flash-Lite: Designed for cost-efficiency, Gemini 3.5 Flash-Lite operates at a speed of 350 output tokens per second. Its pricing is set at $0.30 per million input tokens and $2.50 per million output tokens, making it a more economical option for specific applications.

Conclusion

The launch of Gemini 3.5 Flash Cyber represents a notable advancement in the application of AI to cybersecurity. Its targeted design for vulnerability detection and patching, coupled with its proven ability to identify critical flaws in the V8 JavaScript Engine, positions it as a significant tool for governments and trusted partners. The simultaneous release of Gemini 3.6 Flash and Gemini 3.5 Flash-Lite further diversifies Google's AI offerings, providing a range of models tailored for different performance and cost requirements.

Sources

• https://deepmind.google/blog/introducing-gemini-3-5-flash-cyber/
• 3.6 Flash, 3.5 Flash-Lite, and 3.5 Flash Cyber
• https://www.theverge.com/tech/968572/google-gemini-flash-cyber-ai-security-model
• Google releases three new Gemini models — but no 3.5 Pro
• With Gemini 3.5 Flash, Google bets its next AI wave on...