Models

Google Launches Gemini 3.8 Flash and Cyber Models

Google has introduced Gemini 3.8 Flash and Gemini 3.8 Flash Cyber, offering developers and security teams advanced reasoning and coding capabilities at a highly competitive price point.

DeepMind Blog14 hrs agoModels
Image: DeepMind Blog

Google has expanded its AI lineup with Gemini 3.8 Flash and Gemini 3.8 Flash Cyber, marking its third Flash release in six weeks. Designed for agentic workflows, Gemini 3.8 Flash is available at an introductory rate of $0.75 per million input tokens and $3.75 per million output tokens, matching Gemini 3.7 Flash. This pricing remains active until December 31, 2026, after which rates rise to $1.50 per million input tokens and $7.50 per million output tokens starting January 1, 2027.

The standard Gemini 3.8 Flash model targets long-horizon software engineering and autonomous agents. It achieves a 54.9% score on the HLE-Verified benchmark and outperforms Gemini 3.7 Flash on the Vals Finance Agent V2 and Harvey's Legal Agent Benchmark. On the DeepSWE v1.1 benchmark, it surpasses larger, more expensive frontier models. For practitioners, this translates to a model designed to invest more computational effort on complex tasks, performing additional reasoning steps and iteratively calling tools. Developers can still opt for lower effort levels or stick with Gemini 3.7 Flash to minimize token overhead.

For security professionals, Gemini 3.8 Flash Cyber delivers specialized vulnerability detection and patching. On the CyberGym benchmark, it outperforms Gemini 3.5 Flash Cyber and larger frontier models. In internal testing across 20 programming languages, the model achieved a success rate exceeding 70%. On the external CWE-Bench patching benchmark, it recorded a pass@1 rate of 47.2%, nearly matching a leading frontier model's 47.8% at a much lower cost. It also features enhanced prompt injection robustness verified by Gray Swan.

Google is already deploying the cyber variant internally. The Chrome Security team reported that the cyber variant generated 2.6 times as many correct vulnerability fixes as larger commercial alternatives, while Google's Cloud Vulnerability Research team used it to find a critical vulnerability in under two hours. Additionally, Wiz observed a 7.5% to 9.7% increase in recall on its internal penetration testing benchmark at a 2.3 to 5.2 times lower cost. While developers can access Gemini 3.8 Flash via Google AI Studio, the Cyber variant is restricted to trusted defenders through the new Fairwind Program.

This is our own summary of reporting by DeepMind Blog

More in Models