Google announces Gemini 3.8 flash models for coding and cybersecurity
Google just dropped two new artificial intelligence models, Gemini 3.8 Flash and Gemini 3.8 Flash Cyber, to make coding and cybersecurity smarter and faster.
Announced today, this is their third "Flash" release in six weeks, and it keeps the same affordable pricing as before: $0.75 per million input tokens and $3.75 per million output tokens.
Gemini Cyber scores 86.2% on CyberGym
Gemini 3.8 Flash has been crushing software engineering benchmarks, solving tough problems better than most larger frontier models, and scoring well on STEM reasoning tests.
The Cyber version is all about finding security flaws: it patches bugs way more accurately (2.6 times better) than larger models in evaluations involving real Chrome security bugs, nails the CyberGym benchmark with an 86.2% score, and spots vulnerabilities across 20 programming languages with more than 70% success on Google's internal benchmark.
Google's new Fairwind program will let government agencies use these tools to protect vital infrastructure, showing they're serious about using artificial intelligence to keep things secure on a national level.