Google launches Gemini 3.8 Flash and Cyber version, delivering software efficiency and vulnerability remediation behind strict testing gates

Google announced the launch of the Gemini 3.8 model family, which includes the Gemini 3.8 Flash model aimed at software development and complex reasoning tasks, and the Gemini 3.8 Flash Cyber model specialized in discovering software vulnerabilities and building effective patches. This offering comes three weeks after the launch of Gemini 3.7 Flash and represents the third Flash release within six weeks, in a rapid pace to update reasoning capabilities and lower operating costs, coinciding with leadership changes at DeepMind that saw Demis Hassabis move to chair of the board and Koray Kavukcoglu assume senior vice-president leadership of the organization.

Google kept the price of the Gemini 3.8 Flash model at $0.75 per million input tokens and $3.75 per million output tokens.with enhanced capabilities in smart-agent tasks and software engineering. The model recorded 59 points on the Artificial Analysis benchmark for intelligence, three points higher than the previous version, and achieved 54.9 percent on the HLE-Verified test, which measures multi-step reasoning accuracy across scientific, professional and humanitarian domains.

In the security track, the Flash Cyber model outperformed the CyberGem benchmark for vulnerability detection, surpassing its predecessor Gemini 3.5 Flash Cyber and larger competing models. It also achieved a 47.2 percent accuracy on the CWE-Bench automated remediation test conducted by Collinear on the first attempt, close to the leading model’s result of 47.8 percent but with lower operating cost. Chrome security team evaluations showed the model produced correct patches at a rate 2.6 times higher than commercial competitors, while Wiz reported a 7.5 percent to 9.7 percent increase in vulnerability-retrieval accuracy in penetration tests at half to one-fifth the cost, and the cloud-vulnerability research unit used it to uncover an architectural flaw in less than two hours compared with months in the usual workflow.

The company limited access to the Flash Cyber model and restricted it to the Fairwind program dedicated to vetted governments, critical-infrastructure operators and approved software developers.This measure reflects a move to separate automated cyber-defense tools from open availability, in anticipation of risks arising from large-scale automated vulnerability discovery and patch generation.

The shift has direct implications for organizations operating in the Gulf, Egypt and the broader Arab region. For tech firms and banks, stable low-cost inference pricing helps reduce expenses for automating software-development cycles and code review without draining compute budgets. Conversely, restricting advanced security models behind audit programs such as Fairwind means that financial-infrastructure operators and regional security operation centers will need to establish formal compliance pathways to obtain these tools, increasing reliance on specialized expertise in managing security-audit platforms rather than on generic, direct APIs.