Google DeepMind has unveiled Gemini 3.5 Flash Cyber, an artificial intelligence model specifically engineered for the discovery and remediation of software vulnerabilities. This specialized AI, built upon the existing 3.5 Flash architecture, is designed to identify, validate, and patch security flaws.
The model will not be made available to the general public due to its "dual-use nature," meaning its capabilities could be applied for both defensive and offensive security operations. Instead, Gemini 3.5 Flash Cyber will be exclusively offered to governments and "trusted partners" through CodeMender, as part of a limited-access pilot program. This restricted deployment aims to give "frontline defenders" an advantage in finding and fixing critical vulnerabilities before they can be exploited.
Google DeepMind emphasized that the model's design prioritizes both speed and cost-efficiency, in addition to accuracy. Its performance and efficiency make it suitable for detecting, validating, and patching code security issues at scale. Gemini 3.5 Flash Cyber is fine-tuned for cybersecurity vulnerability work and offers a lower cost per token compared to larger AI models. Within CodeMender's multi-agent setup, where multiple 3.5 Flash Cyber agents operate in parallel, this cost advantage is amplified, leading to greater code coverage per session.
The company stated that the model achieves competitive performance on the CyberGym benchmark, a standard evaluation for this class of AI capability. Google's announcement acknowledges that AI models are increasingly capable of finding security vulnerabilities faster than current systems can address them, necessitating a highly capable and efficient approach to software security. DeepMind also confirmed future plans to incorporate red-teaming features and end-to-end enterprise defense capabilities into the model over time.
Alongside Gemini 3.5 Flash Cyber, Google also released two other models: Gemini 3.6 Flash and Gemini 3.5 Flash-Lite. Gemini 3.6 Flash is an updated general-purpose model, offering improved performance in coding, knowledge work, and multimodal tasks. It consumes 17% fewer output tokens than 3.5 Flash, according to the Artificial Analysis Index, and is priced at $1.50 per million input tokens and $7.50 per million output tokens. This enhanced efficiency and lower cost aim to make agentic tasks more cost-effective. On benchmarks like DeepSWE, it shows up to a 65% token reduction in some configurations and outperforms 3.5 Flash in software engineering (49% vs. 37% on DeepSWE), machine learning research (63.9% vs. 49.7% on MLE Bench), and computer use tasks (83.0% vs. 78.4% on OSWorld-Verified).
Gemini 3.5 Flash-Lite is designed for speed and scalability, operating at 350 output tokens per second, as measured by Artificial Analysis. It is priced at $0.30 per million input tokens and $2.50 per million output tokens, offering a strong price-to-performance ratio for high-throughput production traffic. This model also includes computer use as a built-in tool and surpasses the older 3 Flash model on several agentic benchmarks.
Both Gemini 3.6 Flash and 3.5 Flash-Lite are immediately available through Google AI Studio, the Gemini API, the Gemini Enterprise Agent Platform, and the Gemini app. Google also noted that Gemini 3.5 Pro is currently undergoing testing with partners and will be broadly released when ready, with pre-training for Gemini 4 already underway.
