What changed
Google shipped Gemini 3.6 Flash on 21 July 2026 at $1.50 per million input tokens and $7.50 per million output tokens, live the same day in Google AI Studio, Android Studio, Google Antigravity, and the Gemini app 1. Google’s own metric is 17% fewer output tokens than Gemini 3.5 Flash to finish the same work, measured on the Artificial Analysis Index 1. Fello AI reports the $7.50 output price is down from $9.00 on 3.5 Flash, with input unchanged at $1.50, and the knowledge cutoff moving from January 2025 to March 2026 2.
Where the savings actually land
The interesting part is not the sticker cut; it is that two levers stack. A lower output price and 17% fewer output tokens per task multiply instead of substituting. Run the arithmetic over the two cited figures: 7.50/9.00, times the 17% token reduction, is 0.833 x 0.83, about 0.69. On output-heavy work the real per-task cost falls by roughly a third, not by the ~17% the price line alone suggests.
| Per 1M tokens | 3.5 Flash | 3.6 Flash |
|---|---|---|
| input | $1.50 2 | $1.50 1 |
| output | $9.00 2 | $7.50 1 |
That number only holds when output dominates the bill.
Impact on your team
If you route high-volume agent traffic through the Flash tier, this is worth re-pricing this week, because the win is concentrated exactly where agent loops spend: output. It is opt-in, not a deprecation, so there is no deadline and no forced migration; 3.6 Flash is available today with no waitlist 1. Two cautions before you flip production. No quality benchmark for 3.6 Flash was captured here, and Google leads with cost and lower latency rather than a task-accuracy delta, so validate on your own evals before assuming 3.5 Flash parity. And the longer knowledge cutoff, March 2026 2, changes what the model knows without you touching a prompt, which can shift behavior on recency-sensitive tasks. Re-price now, re-benchmark before you commit.