Google unveils gemini 3.8 flash and cyber models with enhanced cybersecurity and coding capabilities

Google has launched Gemini 3.8 Flash and 3.8 Flash Cyber, two new models designed to enhance reasoning, coding, and cybersecurity capabilities. The 3.8 Flash variant outperforms previous versions in complex engineering tasks, achieving a 54.9% score on HLE-Verified and surpassing larger frontier models on DeepSWE v1.1. It also excels in benchmarks like Vals Finance Agent V2 and Harvey's Legal Agent Benchmark. The model’s performance gains stem from its ability to execute extra reasoning steps and call tools iteratively, especially at higher effort levels. Developers can opt for lower effort levels to minimize token overhead or continue using Gemini 3.7 Flash for efficiency-first workloads. Gemini 3.8 Flash Cyber, available through the Fairwind Program, demonstrates frontier-level performance in autonomous vulnerability discovery on CyberGym and a 70% success rate on an internal benchmark covering 20 programming languages. It focuses on vulnerability fixing over offensive capabilities, achieving a 47.2% pass@1 on CWE-Bench, outperforming a leading frontier model at 47.8% while being more cost-effective. The model includes safeguards against misuse in CBRN and cyber offense domains, with a more permissive set of mitigations for cybersecurity.