Summary
On September 3, 2026, Google released Gemini 3.8 Flash to all developers alongside Gemini 3.8 Flash Cyber, a gated variant reserved for vetted security teams under its Fairwind program. Google says the third Flash release in six weeks makes sizable gains on software engineering, agentic work, and multi-step reasoning, and on DeepSWE v1.1 it beats most larger frontier models at solving complex engineering problems end to end. Launch pricing matches Gemini 3.7 Flash at $0.75 per million input tokens and $3.75 per million output tokens.
What changed
Google released Gemini 3.8 Flash via the Gemini API, AI Studio, Android Studio, Stitch, Gemini Enterprise, and consumer surfaces (Search, Sheets), with a gated Gemini 3.8 Flash Cyber variant for government, critical-infrastructure operators, and software maintainers that Google claims produced 2.6x more correct patches than the best commercial models. The standard tier keeps the 3.7 Flash price and includes CBRN and cyber-offense safeguards; 3.7 Flash remains available as a lower-cost option.
Why it matters
Google is compressing frontier coding and agentic performance into its cheapest tier at the pace of a monthly refresh, keeping Flash's price flat while claiming DeepSWE wins over larger models — meaning agent builders and CI-time coders can trade up without redoing budgets. The gated Cyber variant is also the first Flash-class model explicitly split into a defender-only SKU, extending the pattern OpenAI set with Astra to a cost-optimized model family.
Evidence excerpt
Sundar Pichai said this is our 3rd Flash release in just 6 wks, noting sizable gains over 3.7 Flash in software engineering, agentic work, and multi-step reasoning. On DeepSWE v1.1, Google says it beats most larger frontier models... Launch pricing matches Gemini 3.7 Flash: $0.75 per million input tokens and $3.75 per million output tokens. Gemini 3.8 Flash Cyber... reserved for vetted security teams through the Fairwind program... produced 2.6 times more correct patches than the best commercial models.