Google introduced Gemini 3.8 Flash on 2 September, its third Flash-family release in six weeks. The general model is available through the Gemini API, AI Studio, selected enterprise services and Pro and Ultra consumer plans. Introductory pricing is $0.75 per million input tokens and $3.75 per million output tokens; Google says both rates will double from 1 January 2027.

Google positions the model for long-horizon coding and agentic tasks. It reports 54.9% on HLE‑Verified and a strong result on DeepSWE v1.1. The company also warns that at higher effort the model performs more reasoning steps and tool calls, so it may consume more tokens. In an independent overview, Ars Technica noted that gains are modest on some tests and that computer-use performance still trails some competitors.

Gemini 3.8 Flash Cyber shares the same foundation but is tuned for vulnerability discovery and patching. Google reports 47.2% on the external CWE‑Bench, versus 47.8% for the leading model. On its own benchmark spanning twenty programming languages it claims more than 70%, while the Chrome team reportedly obtained 2.6 times more correct patches than with the best larger commercial models. The last two figures are internal or partner measurements.

The Cyber variant is not a public offensive-security tool. Google is offering it to government bodies, critical-infrastructure operators and software maintainers through the vetted Fairwind program because it has more permissive cyber mitigations than standard Flash. Its practical value may be faster bug triage and patch proposals, but every fix still needs a reproducible test, human review and controlled deployment.