Google has added another rapid advancement to its artificial intelligence efforts by announcing the new generation models Gemini 3.8 Flash and the cybersecurity-focused Gemini 3.8 Flash Cyber. Coming just three weeks after the introduction of Gemini 3.7 Flash, this leap promises significant performance improvements, particularly in autonomous software development, multi-step logical reasoning, and cyber defense. Moreover, Google is making the new models available to developers with the same affordable price and high speed advantages as the previous version.

Powerful Workforce for Developers: Gemini 3.8 Flash

Gemini 3.8 Flash is positioned as the most capable version to date in the lightweight and fast Flash series. Google specifically optimizes this model for end-to-end software engineering, autonomous AI agent workflows, and complex enterprise-level data processing. Compared to the previous version, the model applies more reasoning steps in demanding tasks, runs tools sequentially, and performs intermediate checks to reduce the error rate.

The model’s biggest advantage is its high success rate in demanding tasks. According to comparisons, 3.8 Flash outperforms many large flagship models in the long-term software engineering benchmark DeepSWE v1.1. It also demonstrates significant superiority in specialized professional workloads such as financial analysis (Vals Finance Agent V2) and legal text analysis (Harvey’s Legal Agent Benchmark). Furthermore, the model achieves a score of 54.9% in the HLE-Verified test covering science and humanities.

Technical Details and Developer Experience

The basic technical features that will make the developers’ job easier are listed as follows:

• Context Window and Output Capacity: The model maintains a context window of 1 million tokens while offering a maximum output token boost of 64,000.
• Adjustable Thinking Levels: Developers can choose the balance between response time and reasoning depth between “Low”, “Medium” (default) and “High” modes. This creates an infrastructure that provides fast processing for chatbots and in-depth analysis for complex code revisions.
• Access and Pricing Policy: The model is offered at an introductory price of $0.75 for 1 million input tokens and $3.75 for 1 million output tokens until December 31, 2026.

A New Era in Cyber ​​Defense: Gemini 3.8 Flash Cyber

One of the most critical modules of the announcement is Gemini 3.8 Flash Cyber, developed directly for cybersecurity experts and defense groups. This special edition specializes in detecting, analyzing, and automatically patching vulnerabilities in code. To prevent the model from being misused by malicious actors, Google is making Flash Cyber ​​​​available only to verified security organizations, critical infrastructure managers, and software developers under the “Fairwind Program.”

Outperforming even large-scale leading models in the industry-standard security benchmark CyberGym, the version achieves a vulnerability detection success rate of over 70% in extensive security tests covering 20 different programming languages. In terms of automated patch development, it shows a 47.2% success rate in the CWE-Bench test, putting it on par with much more valuable flagship models.

Tested in Real-World Scenarios

Google is also actively deploying the new security model on its own infrastructure. According to shared data, the Chrome Security Team states that Flash Cyber ​​3.8 produces 2.6 times more accurate security patches against known vulnerabilities compared to major commercial models on the market. Similarly, cybersecurity company Wiz reports that in its penetration tests, the model achieved a 7.5% to 9.7% higher vulnerability detection rate, and did so at a cost 2 to 5 times lower. Furthermore, the Google Cloud Security Research Group reports that they detected a critical infrastructure vulnerability that could have taken months to resolve in less than 2 hours using this model.

Gemini 3.8 Flash is available to developers via Google AI Studio, Gemini API, Android Studio, and Google Cloud, while on the end-user side it is offered to Google AI Pro and Ultra subscribers. Google’s reduction of the model development cycle to weeks indicates that the competition for speed and efficiency in artificial intelligence will intensify significantly in the coming period.