Google Gemini 3.8 Flash arrives with huge focus on cybersecurity
What's the story
Google has launched its latest AI model, Gemini 3.8 Flash, with a special focus on agentic workflows and cybersecurity. The new model is the third in the Flash series to be released in just six weeks. It comes in two variants: the standard Gemini 3.8 Flash and a specialized version for cybersecurity, called Gemini 3.8 Flash Cyber.
Developer's delight
Gemini 3.8 Flash outperforms most larger models on DeepSWE v1.1
The standard version of Gemini 3.8 Flash is targeted at developers working on agentic applications and workflows.
It is also available to regular Google AI Pro and Ultra subscribers through the Gemini app.
The new model has proven its mettle by outperforming most larger frontier models on the DeepSWE v1.1 benchmark, which tests autonomous complex engineering problem-solving capabilities, scoring an impressive 73.7%, just below Claude Opus 5's 74% score.
Performance metrics
'3.8 Flash works harder,' says Google
The new model also outperformed its predecessors on several other benchmarks, including financial analysis, complex legal workflows, information synthesis, agentic terminal coding, and long video understanding.
Google said that "3.8 Flash works harder," meaning it shows more diligence on complex tasks by taking additional reasoning steps when needed.
This could mean the model may sometimes use more tokens to ensure maximum performance.
Cost considerations
Pricing and competition with other AI models
The new model is priced at $0.75 per million input tokens and $3.75 per million output tokens as part of an introductory offer.
However, these prices will be revised to $1.50 and $7.50, respectively, in the future.
Despite this, the model is expected to be cheaper than its competitors such as Claude Opus 5 and GPT-5.6 Sol in the long run.
Cybersecurity focus
Gemini 3.8 Flash Cyber aimed at government authorities, software maintainers
The second variant, Gemini 3.8 Flash Cyber, is a specialized cybersecurity model that comes with advanced vulnerability detection and automated patching capabilities.
However, access to this powerful tool is limited through Google's Fairwind Program for trusted government authorities, critical infrastructure operators, and software maintainers only.
On the industry-standard benchmark for finding vulnerabilities in C/C++, this model scored an impressive 86.2%, outperforming its predecessor and larger models like Mythos 5 and GPT-5.5-Cyber.
Patching prowess
Automated patching capabilities for cybersecurity defense
The standout feature of Gemini 3.8 Flash Cyber is its automated patching ability.
The model can generate patches to fix vulnerabilities, making it a valuable tool for defenders against cyber threats.
According to Google, the model produced 2.6 times more correct patches for vulnerabilities in Chrome than "the best commercial models that are much larger."