Google Updates Gemini Flash Again With 3.8 Flash and Cyber

Google recently announced Gemini 3.8 Flash, less than a month after releasing Gemini 3.7 Flash.

The update also introduced Gemini 3.8 Flash Cyber, the first update to Google’s cybersecurity-specific Flash model since the company introduced three Flash models in July. 

Once again, the improvements push Flash forward with stronger reasoning, better performance on agentic tasks, and expanded cybersecurity capabilities.  

What’s New With the Two Gemini 3.8 Flash Models 

Google’s benchmark results showed significant progress, but the most impressive results were concentrated in specialized areas. 

Larger frontier models still hold an advantage on some complex agentic workflows, but 3.8 Flash makes significant leaps without giving up the speed, cost, and efficiency that make the series popular.

Updates to Gemini 3.8 Flash 

Gemini 3.8 Flash makes its biggest gains in coding and technical use cases. Google shared strong results in long-horizon software engineering, multi-step problem-solving, and sustained agentic work.

Key updates included: 

• Long-horizon coding: One of the standout results was DeepSWE v1.1, where Gemini 3.8 competed and even outperformed larger frontier models on complex software engineering tasks.
• Research and analysis: 3.8 Flash posted stronger results across finance, legal, and scientific reasoning.
• Video understanding: The model improved to 87.8% on agentic long video understanding on LVBench.
• Technical agent work: On Terminal-Bench 4.0, Gemini 3.8 Flash scored 19.1%, which was up from 11.2% on 3.7 Flash.

The benchmark picture for Gemini 3.8 Flash isn’t consistently better than frontier models across the board, but its performance on coding and industry-specific work is eye-opening. This is especially notable given that input and output token costs remained unchanged from 3.7 Flash.

Restricted Release of Gemini 3.8 Flash Cyber 

Gemini 3.8 Flash Cyber doubles down on one of the Flash lineup’s clearest strengths, which is specialized work. The model is built for defensive cybersecurity, with a focus on locating vulnerabilities and automating patching.

The model was released through the Fairwind Program, which is designed for authorized defenders using advanced AI systems for cybersecurity.

Areas of strength: 

• Vulnerability detection: Higher performance on CyberGym than some larger models 
• Agentic patching: Scored a 47.2% pass rate on CWE-Bench, close to results from frontier models 
• Faster response: Security teams can quickly move from identifying to fixing security concerns

Google also tested Flash Cyber across 20 programming languages, which signals flexibility across different codebases. 

Staying Up to Speed With Specialized AI 

Gemini 3.8 Flash continues to make major strides in coding and other advanced agentic tasks while maintaining the low introductory pricing of 3.7 Flash.

What brands can’t afford to ignore is how quickly AI models are improving, and how the speed and cost of operating models such as Gemini 3.8 Flash can make advanced capabilities accessible to more customers.

With agentic workflows becoming more capable of handling coding, scientific research, and other technical tasks, companies have more opportunities to build faster, more responsive operations using the latest AI technology.

Learn more about using AI innovation and human expertise to move faster, adopt new technology first, and set the pace for your industry.