Quick answer

Google released Gemini 3.8 Flash on September 2, 2026 — its third Flash-tier model in six weeks. Pricing is an introductory $0.75 per million input tokens and $3.75 per million output tokens through December 31, 2026, rising to $1.50 and $7.50 on January 1, 2027. The context window is 1,048,576 tokens and Google reports 54.9% on HLE-Verified. A second variant, Gemini 3.8 Flash Cyber, ships with looser safety mitigations to governments and infrastructure operators through Google's Fairwind Program; its pricing and availability are not public.

Flash is the Gemini tier most people actually use: it powers the free Gemini app, Google Workspace features, and a large share of API traffic because it is fast and cheap. Three Flash releases in six weeks tells you Google is iterating on its volume model faster than on its flagship — and the pricing tells you it is fighting for developers who might otherwise default to a Chinese model or GPT-5 mini.

What changed

  • Introductory pricing of $0.75 input / $3.75 output per million tokens until December 31, 2026
  • Standard pricing from January 1, 2027: $1.50 / $7.50 — a doubling that developers should budget for now
  • Context window of 1,048,576 tokens, the full million
  • 54.9% on HLE-Verified (Humanity's Last Exam), Google's headline reasoning number
  • Flash Cyber: a variant with reduced safety mitigations for approved government and critical-infrastructure users

The pricing trick to watch

Introductory pricing is new for Google at this scale. It means every cost comparison you read this autumn is temporarily flattering to Gemini, and every product built on it will see costs double on New Year's Day. If you are choosing a model for something you will still run in 2027, compare against the January prices, not the September ones. At $1.50/$7.50 Flash is still competitive, but Qwen3.8-Omni-Flash at $0.15/$0.47 and Grok 4.7 at $2/$6 change the picture.

What Flash Cyber signals

A version of the model with looser guardrails, available only to vetted governments and infrastructure operators, follows the pattern Anthropic set with Mythos and OpenAI with Astra's cybersecurity safeguards. The frontier labs have converged on the same answer to dangerous capability: ship it, but only to people you have checked. Whether that is responsible governance or a two-tier internet depends on who you ask; either way it is now the industry standard.

Related reading

Gemini 3.5 Flash Launch — Google's Fast, Cheap Model What Is Google Gemini? AI API Pricing in 2026 — The Honest Comparison September 2026 in AI — Eight Frontier Launches Explained

Bottom line

Gemini 3.8 Flash is a strong, cheap, huge-context model with a price that doubles in January. Use it now, model your 2027 costs at the standard rate, and treat the Cyber variant as the clearest sign yet that "frontier capability" now comes in a public version and a vetted one.