On September 2, 2026 Google unveiled Gemini 3.8 Flash Cyber, a cybersecurity-tuned brother to its new general-purpose Flash 3.8, and debuted a new limited- membership trusted-defenders program instead of launching it widely. The company claimed the duo were its third Flash-family refresh in just six weeks and stated that 3.8 Flash and the Cyber version have shared DNAsimply that the latter is tuned to detect and patch software flaws in Flash speed and at cost.
Gemini 3.8 Flash Cyber is the strongest cybersecurity model Google has built to date, as Google, with frontier-equivalent results in autonomous vulnerability discovery and auto-patching. On Cyber Gym, the industry standard for vulnerability discovery, it beat even the much larger frontier models and the prior Gemini 3.5 Flash Cyber.
On an internal Google benchmark addressing difficult code for 20 languages, it beat a 70+ percent success rate. On Collinear’s CWE-Bench patching test, it beat a frontier model with 47.8 percent pass@1 at 47.2 percent. Google claimed it focused its efforts on defense rather than offense or exploitation during development of the model. It can explore even large codebases, bring hidden problems to the surface, produce validated patches, and enable protection teams to close exploitation windows.
Inside, Google’s Chrome security team managed 2.6 more correct patches than with the much bigger commercial models tested. A Cloud partner said the model uncovered a key bug in around 2 hours, a months-long task under normal circumstances.
Partners like Wiz and Palo Alto Networks also reported superior recall to real bug data sets while costing less. Access is limited with the Fairwind Program, announced the same day, targeting government agencies, operators of critical infrastructure, software maintainers, Google Cloud customers, and specific cybersecurity partners like CrowdStrike and the Center for Internet Security. TheCybermodel, which is used with a more liberal set of cybersecurity mitigations than the publicly facing Gemini 3.8 Flash, is not being released to the open API. The Fairwind program also combines the model with CodeMender, Google’s bug-hunting and remediation agent.
The broader Gemini 3.8 Flash offering meanwhile is also launching immediately on the Gemini API, Google AI Studio, Android Studio, Gemini Enterprise, and consumer interfaces for the AI Pro and Ultra tiers. Google set its price at the same introductory rate as 3.7 Flash, $0.
75 per million input tokens and $3.75 per million output tokens through the end of 2026. It said 3.8 Flash “performs more work” by doing more reasoning steps and iterative tool use on agentic and software-engineering tasks, and it posted improvements across coding and agent benchmarks while maintaining Flash-tier latency.
Poland’s twin unveilings also reveal the degree to which Google is dividing the single model family into one massively accessible workhorse and a aggressively protected roaming defender. To the not-Java developer, it’s a cheaper, speedier code-autonomous agent story. To administrations and network operations Faireswleded into… Gemini 3.8 Flash Cybermas it’s presented as an accessible hardware for closing the elapsed time.
Digital entrepreneur and content expert I help businesses with AI, SEO and the latest tech trends. I started Silicon Valley Weekly to make complex tech concepts easy to understand and use for business growth. I know a lot about systems and help startups, entrepreneurs and brands navigate the fast-changing world of tech and online marketing.
I build strategies that use data, search optimization, content marketing and AI tools to get visibility, engagement and revenue. I love finding ways for businesses to grow increasing their presence and turning new ideas into successful businesses. My goal is to connect the technology, with practical business use so brands can succeed online.
