Google DeepMind has released its latest Gemini models, Gemini 3.8 Flash and Gemini 3.8 Flash Cyber.
These models are designed to be optimized for agentic workflows and cybersecurity.
Gemini 3.8 Flash maintains the same speed and low cost as 3.7 Flash while improving reasoning and coding performance.
According to the company, it demonstrated performance exceeding larger frontier models in the DeepSWE v1.1 benchmark.
Furthermore, it reportedly outperforms 3.7 Flash and other frontier models in specialized domain benchmarks such as Vals Finance Agent V2 and Harvey's Legal Agent Benchmark.
The models are equipped with mechanisms to process complex tasks more meticulously by executing additional reasoning steps or making iterative tool calls.
Gemini 3.8 Flash Cyber will be provided to trusted defenders through the Fairwind Program.
Google DeepMind stated that it has demonstrated performance exceeding 3.5 Flash Cyber and large-scale frontier models in the CyberGym benchmark, which evaluates vulnerability discovery.