Google unveils new Gemini AI models, including cybersecurity-focused version

The latest Gemini releases highlight Google's push to strengthen AI-driven software development and cybersecurity, with a restricted security model for critical vulnerability detection alongside faster, more efficient and lower-cost models for developers and enterprises.

Google has introduced three new additions to its Gemini artificial intelligence portfolio, including a specialised cybersecurity model aimed at helping organisations identify and fix software vulnerabilities. While two of the models have been made available to developers and users, the cybersecurity-focused version will initially be accessible only to governments and a select group of trusted partners because of its sensitive capabilities.

The new cybersecurity model, called Gemini 3.5 Flash Cyber, is built on the Gemini 3.5 Flash architecture and has been trained specifically for software security tasks such as detecting, validating and remediating vulnerabilities in code. The model powers Google's AI-driven CodeMender platform, which was introduced in 2025 to automate vulnerability discovery and software patching. By repeatedly analysing different code paths, the system is designed to improve coverage and compile its findings into a consolidated report.

Google said the model combines competitive performance with lower operating costs compared to larger cybersecurity-focused AI systems. According to the company, testing on the CyberGym benchmark a framework used to evaluate AI agents against real-world software vulnerabilities showed strong results. During assessments involving Google's V8 JavaScript Engine, Gemini 3.5 Flash Cyber reportedly identified 55 confirmed vulnerabilities, outperforming Gemini 3.5 Flash and Anthropic's Claude Opus 4.6. Google also said the model detected several vulnerabilities that competing systems did not identify.

Explaining the restricted rollout, Raluca Ada Popa, Gemini Security Lead at DeepMind, and Four Flynn, Vice President of Security and Privacy, said: "Given the dual-use nature of this technology, we have taken an intentional approach to how we deploy 3.5 Flash Cyber. This will give frontline defenders a head start in finding and fixing critical vulnerabilities before they can be exploited, while mitigating against broader misuse."

Google expands Gemini portfolio with faster, more efficient models

Alongside the cybersecurity model, Google announced Gemini 3.6 Flash and Gemini 3.5 Flash-Lite, both of which are now available to the public through the Gemini app, Google AI Studio and Android Studio.

According to the company, Gemini 3.6 Flash delivers stronger coding and multimodal capabilities while requiring fewer output tokens to complete tasks than its predecessor, making it more efficient and cost-effective. The model is priced at $1.50 per million input tokens and $7.50 per million output tokens, reflecting a reduction in output costs compared with Gemini 3.5 Flash.

Gemini 3.5 Flash-Lite has been positioned as the fastest and most economical model in the 3.5 family. Google said it can generate output at speeds of up to 350 tokens per second and is intended for applications requiring high throughput at lower cost.

Focus on AI security and future development

The launch reflects Google's continued efforts to strengthen its position in the rapidly evolving AI market, particularly in software development and cybersecurity. Industry observers have noted that the cybersecurity model could help Google compete more aggressively in AI-assisted code security, an area where rivals have also been investing.

Google also confirmed that Gemini 3.5 Pro remains in partner testing ahead of a broader release. At the same time, the company said development of Gemini 4 is progressing, describing it as its most ambitious AI pretraining effort to date.