What happened: Google released Gemini 3.8 Flash on September 2 as its most capable Flash-tier model for long-horizon software engineering, autonomous agents, and multi-step reasoning. A parallel variant, Gemini 3.8 Flash Cyber, is available only through the new Fairwind Program for trusted defenders.

The general model keeps the introductory price of $0.75 per million input tokens and $3.75 per million output tokens through year-end. Context window sits at 1,048,576 tokens. Google reported gains over 3.7 Flash on Terminal-bench, SWE-Bench Pro, and internal long-horizon coding suites. The Cyber variant scored 86.2% on CyberGym for vulnerability discovery.

Primary sources: Google’s launch post, the Gemini API model card, and partner statements from Chrome Security and Wiz.

Key takeaways

What the Flash-Cyber split actually means

The architecture is shared; the difference is access and safety posture. The public model is positioned as a workhorse for coding agents and enterprise workflows. The Cyber variant is tuned for vulnerability detection and automated patching and is restricted to approved government, critical-infrastructure, and software-maintainer accounts.

This mirrors the pattern already visible at OpenAI with Astra and at Anthropic with Mythos-class models: high cyber capability is released under controlled programs rather than open API access. For builders using the AI content stack, the practical implication is clearer model selection — use the general Flash for production agents, and treat the Cyber line as a specialized defensive tool rather than a general coding assistant.


What to watch next

Watch Fairwind enrollment criteria and any expansion beyond the initial defender cohort. Watch independent benchmark runs against Claude Fable 5.1 and GPT-class models on the same agentic coding suites. Watch whether Google ships a Pro-tier update or continues the rapid Flash cadence. The release is live; the governance experiment is just starting.