Google DeepMind releases Gemini 3.8 Flash and 3.8 Flash Cyber with improved reasoning, coding, and cybersecurity capabilities

Google DeepMind introduced Gemini 3.8 Flash and 3.8 Flash Cyber, the latest variants in its Flash series, building on the 3.7 Flash release from three weeks prior. The company describes 3.8 Flash as its most intelligent "workhorse" model, delivering significant improvements over 3.8 Flash in software engineering, agentic tasks, and multi-step reasoning across specialized domains.

Gemini 3.8 Flash is positioned as a cost-effective upgrade, available at the same introductory price as 3.7 Flash: $0.75 per million input tokens and $3.75 per million output tokens. The model is designed for long-horizon coding and autonomous agents, with Google highlighting its ability to approach the performance of higher-cost frontier models on complex engineering tasks. On the DeepSWE v1.1 benchmark for long-horizon software engineering, 3.8 Flash outperforms most larger frontier models while maintaining a fraction of their cost.

The model's improvements stem from a design choice emphasizing diligence on complex tasks, including executing extra reasoning steps and iterative tool use. For applications prioritizing compute efficiency, developers can opt for lower effort levels to reduce token overhead, while 3.7 Flash remains supported for efficiency-first workloads. Google also showcased demonstrations of 3.8 Flash's capabilities, such as building a fully functional DOS version of Google Maps in a single prompt within Google Antigravity.

Gemini 3.8 Flash Cyber, available to trusted defenders through the Fairwind Program, targets autonomous vulnerability detection and automated patching. Google claims frontier-level performance on the CyberGym benchmark, surpassing both 3.5 Flash Cyber and significantly larger frontier models. The model was evaluated against an internal benchmark to assess real-world defensive needs beyond C/C++ codebases, reflecting a broader scope of cybersecurity applications.

Both models share a foundational intelligence and are accelerated by long-running agentic loops that recursively evaluate and refine outputs. Google DeepMind's announcement emphasizes the practical deployment of these models in enterprise and cybersecurity contexts, where reliability and cost efficiency are critical factors.