Gemini 3.8 Flash Launch Pricing Ends December 31: API Price Doubles in 2027
Google made Gemini 3.8 Flash generally available in the Gemini API on September 2, 2026. The release notes call it Google’s most intelligent Flash model, aimed at long-horizon software engineering, autonomous agents and complex enterprise workflows. Its current API price is temporary.
Now and from January 1, 2027
Paid tier, standard processing, text input, per 1M tokens, from Google’s pricing page as checked on September 16, 2026:
| Gemini 3.8 Flash | Input | Context caching | Output |
|---|---|---|---|
| Through December 31, 2026 | $0.75 | $0.075 | $3.75 |
| From January 1, 2027 | $1.50 | $0.15 | $7.50 |
Every rate doubles. A request with 1,000 input and 500 output tokens goes from about $0.0026 to $0.0053. Context caching also charges storage at $1.00 per 1M tokens per hour. A free tier is available.
Other Flash models on launch pricing
Gemini 3.8 Flash is not the only model with a deadline:
- Gemini 3.6 Flash, generally available since July 21, 2026, has the same prices and the same schedule: $0.75 / $3.75 now, $1.50 / $7.50 from January 1, 2027.
- Gemini 3.7 Flash, generally available since August 13, 2026, also costs $0.75 input and $3.75 output at an introductory price through December 31, 2026. Google’s pricing page does not yet state its 2027 price.
For comparison, Gemini 3.5 Flash is not on a promotion and costs $1.50 input and $9.00 output. Even after the increase, Gemini 3.8 Flash stays cheaper than 3.5 Flash on output.
What to do before January
Budgets built on today’s Flash prices will be off by a factor of two from the first day of 2027. If you are estimating annual costs, use the 2027 rates. If your workload tolerates delay, check Google’s pricing page for batch rates: on the Gemini models where Google lists them, batch costs half the standard price.
All Gemini rates, including long-context tiers for Gemini 3.1 Pro and 2.5 Pro, are on our Google Gemini API pricing page. The full list of dated price changes is in the model changelog.