Google AI: Flash Models Advance, Pro Stalls

Google has unveiled its newest AI offerings, Gemini 3.8 Flash and Gemini 3.8 Flash Cyber, marking a rapid succession of model releases. The Gemini 3.8 Flash model is positioned as a versatile tool, adept at agentic tasks and software development. Its companion, Gemini 3.8 Flash Cyber, has been specifically tuned for detecting and mitigating cybersecurity vulnerabilities.

These models are now accessible to developers via API, with introductory pricing available through the end of 2026. Google presents these new Flash models as its most capable reasoning and coding tools to date. Benchmark results indicate that Gemini 3.8 Flash leads on the DeepSWE leaderboard for complex software engineering problems.

While Gemini 3.8 Flash shows promise in areas like coding, its performance in agentic computer use, as measured by OSWorld-2.0, still trails leading models like Claude Opus. The Gemini 3.8 Flash Cyber model has undergone internal testing, reporting substantial gains in identifying vulnerabilities and issuing patches, with notable improvements seen by the Chrome and Cloud security teams.

Gemini 3.8 Flash is available across Google's ecosystem, requiring a Pro or Ultra subscription for app access, or use via AI Studio. Gemini 3.8 Flash Cyber is currently limited to trusted testers and government entities.