Google Launches Gemini 3.8 Flash and Cyber Models
New AI models enhance coding, reasoning, and cybersecurity at lower costs.
1Introduction to Gemini 3.8 Flash
Google has unveiled Gemini 3.8 Flash and its cybersecurity-focused variant, 3.8 Flash Cyber, marking its third Flash model release in six weeks. The standard 3.8 Flash excels in agentic tasks, software development, and multi-step reasoning, while 3.8 Flash Cyber specializes in vulnerability detection and mitigation. Both models build on the success of 3.7 Flash, offering improved performance at the same low cost. CEO Sundar Pichai highlighted significant leaps in software engineering and reasoning capabilities.
2Performance Benchmarks
Gemini 3.8 Flash outperforms many larger frontier models on the DeepSWE coding benchmark, achieving superior results at a fraction of the cost. It also shines in specialized domains like finance and law, scoring 54.9% on HLE-Verified for multi-step reasoning. Meanwhile, 3.8 Flash Cyber achieved 86.2% on CyberGym and 47.2% on CWE-Bench, showcasing its prowess in vulnerability discovery and patching. Internal tests revealed a 70%+ success rate in detecting vulnerabilities across 20 programming languages.
3Enhanced Reasoning and Diligence
A key design choice for 3.8 Flash is its ability to 'work harder,' executing extra reasoning steps and iteratively calling tools for complex tasks. This diligence sometimes increases token usage but maximizes performance. The model handles text, images, audio, video, and PDFs, with a 1M-token input window and 64K-token output limit. Developers can adjust effort levels to balance quality, cost, and latency, or stick with 3.7 Flash for efficiency-first workloads.
4Pricing and Accessibility
Gemini 3.8 Flash is priced at $0.75 per million input tokens and $3.75 per million output tokens, matching 3.7 Flash's introductory rates. It’s available in Gemini Enterprise and via Google AI Studio, Antigravity, Android Studio, and Stitch for UI generation. Google emphasizes flexibility, allowing users to customize model effort levels based on their needs. Senior product director Tulsee Doshi noted the model’s adaptability for diverse workloads, from high-performance to cost-sensitive applications.
5Cybersecurity Specialization
3.8 Flash Cyber is Google’s most capable cybersecurity model, matching frontier-level performance in vulnerability discovery and patching. Its training in rigorous cybersecurity domains contributes to its shared core improvements with 3.8 Flash. The model’s agentic loops recursively evaluate and refine its underlying intelligence, enhancing reliability for critical enterprise tasks. Google’s blog highlights its dependability in autonomous problem-solving and advanced analysis.
6Developer Flexibility
Developers can choose between 3.8 Flash’s higher performance or 3.7 Flash’s efficiency, depending on their priorities. Lower effort levels minimize token overhead for compute-constrained applications. Google assures continued support for 3.7 Flash, which remains ideal for efficiency-first workloads. The 3.8 models’ multimodal capabilities and long-context handling make them versatile for coding, knowledge work, and scientific reasoning, per benchmark evaluations.
7Industry Impact
The rapid release cycle—three Flash models in six weeks—signals Google’s aggressive AI innovation pace. 3.8 Flash’s ability to rival frontier models at lower costs could disrupt the market, particularly in coding and cybersecurity. Its specialized knowledge domain performance, like legal and financial benchmarks, positions it as a tool for professional fields. The models’ iterative refinement and agentic workflows set a new standard for enterprise AI solutions.
8Future Prospects
Google’s Gemini 3.8 Flash and Cyber models represent a leap in AI-driven efficiency and specialization. Their success in coding, reasoning, and cybersecurity benchmarks suggests broad applicability across industries. With customizable effort levels and competitive pricing, they cater to diverse user needs. As Google continues to refine its Flash lineup, these models could redefine expectations for cost-effective, high-performance AI in enterprise and developer ecosystems.
Original article: VentureBeat — Google Launches Gemini 3.8 Flash and Cyber Models