Introducing Gemini 3.8 Flash and 3.8 Flash Cyber
Google ● Covered by 4 sources
Google DeepMind launched Gemini 3.8 Flash and a cyber version built for defenders. It says the new models keep 3.7’s speed and cost while pushing harder on coding, reasoning, and patching.
Based on reporting by Google — read the original for the full story.
Summary, retelling and take written by AI under human oversight; images are AI-generated illustrations. How we work · Report an error
Google DeepMind is rolling out Gemini 3.8 Flash and Gemini 3.8 Flash Cyber, and the pitch is pretty clear: keep the speed and price of 3.7 Flash, then squeeze a lot more out of the same core. The company says this is its third Flash release in six weeks, building on 3.7 Flash from three weeks ago. Gemini 3.8 Flash starts at $0.75 per million input tokens and $3.75 per million output tokens, the same introductory pricing as 3.7 Flash.
The regular 3.8 Flash is aimed at long-horizon coding and agentic work. Google says it improved across software engineering, multi-step reasoning, and specialized domains, and that on DeepSWE v1.1 it outperforms most larger frontier models at solving complex engineering problems end to end. It also posts a 54.9% score on HLE-Verified, which Google uses to point to broader reasoning across STEM, humanities, and professional fields. The model is built to work harder on tough tasks, with extra reasoning steps and iterative tool use when needed.
The cyber variant is the sharper part of the release. Gemini 3.8 Flash Cyber is only available to trusted defenders through the new Fairwind Program, and Google describes it as its most capable cybersecurity model. On CyberGym, the standard benchmark for vulnerability finding, it reaches frontier-level performance and beats both 3.5 Flash Cyber and larger frontier models. In Google’s own benchmark across codebases in 20 programming languages, it goes past previous models and tops a 70% success rate.
Patch work is the other half of the story. On CWE-Bench from Collinear, 3.8 Flash Cyber posts a 47.2% pass@1, just behind a leading frontier model at 47.8%, while coming in at a much lower cost. Google says it is already using the model internally: Chrome Security found it produced 2.6 times more correct patches than the best commercial models that are much larger, Wiz saw 7.5% to 9.7% better recall at 2.3x to 5.2x lower cost, and Google’s Cloud Vulnerability Research team used it to find a critical foundational bug in under 2 hours.
Both models come with safety measures around CBRN misuse and cyber offense, and Google says Gemini 3.8 Flash also improved on prompt-injection robustness. For builders, the company is pushing 3.8 Flash through Google Antigravity, the Gemini API, Google AI Studio, Android Studio, Stitch, Gemini Enterprise, and the Gemini app. The cyber model is reserved for trusted government authorities, critical infrastructure operators, and software maintainers who can apply for access through Fairwind.
My take — AI-written commentary, not fact-checked reporting
This is Google doing the obvious thing, which is also the smart thing: make the general model cheap, then wall off the dangerous cyber version behind a gate. The open model crowd will complain about access, but nobody serious wants “more capable cyber” handed out like free pizza. The real tell is how much of this launch is about patching, not hacking; that’s the industry finally admitting defense is where the clean money and the cleaner conscience are.
Read more about this at: Google