Gemini 3.7 Flash launches at half price. The other half arrives in January

Google shipped Gemini 3.7 Flash on August 13, three weeks after 3.6 Flash. The benchmark numbers are good. The line worth reading twice is on the pricing page: $0.75 per million input tokens and $3.75 per million output tokens, marked as an introductory rate that ends December 31.

On January 1, 2027, those numbers become $1.50 and $7.50. Double, on a date already printed on the label. If any part of your business runs on per-token AI pricing, whether you know it or not, Gemini 3.7 Flash just handed you a budgeting deadline.

What's new in Gemini 3.7 Flash

Skipping the leaderboard theater, three of the gains matter to a business owner in plain terms.

• Workflow automation nearly doubled: 30.4% on AutomationBench, up from 17% for the previous model. This is the "follow a multi-step process without wandering off" score, the one that decides whether an automated quote-follow-up sequence works or embarrasses you.
• Document comprehension went from 22% to 34% on a PDF-heavy benchmark. Reading supplier contracts, insurance documents, and government forms is a large share of what small businesses ask AI to do.
• Coding scores jumped too (43.6% vs 34.4% on FrontierCode, 65.3% vs 49% on long-horizon engineering tasks), which mostly matters to you through the agencies and freelancers who build on top of it.

For individuals, the model rolls out through Gemini Spark to Google AI Pro and Ultra subscribers in 160+ countries. Developers get it in AI Studio and Antigravity, and enterprises through the Gemini Enterprise platform.

The fine print: a price with an expiry date

Intro pricing on models isn't new, but a printed doubling date is unusually honest, and unusually easy to plan around. Here's who it touches.

If you use the Gemini app on a subscription, nothing changes for you in January. Your $20-30 a month buys whatever model Google serves, and 3.7 Flash arriving inside that subscription is a straight upgrade. Enjoy it.

If you pay per token, directly or through a middleman, the doubling is your problem. That includes Zapier or Make scenarios calling Gemini, a chatbot an agency built you last spring, or any "AI-powered" feature in your stack where the vendor passes model costs through. A workflow that costs $40 a month at intro rates costs $80 in January. Multiply across every automation you've stacked up since spring and the 2027 line item stops being a rounding error.

There's a second-order version of the same problem: an agency that quoted you a flat monthly fee against intro-rate token costs will come back to renegotiate, or quietly cut quality. Worth a question at your next check-in.

What to do before December 31

Three moves, none of them urgent, all of them cheaper now than in January.

1. List what you pay for. Most owners we talk to can't name every AI subscription and API-billed automation they're running. Our AI Spend Auditor skill turns that into a standing inventory with renewal dates and expiry flags, this Gemini date included.
2. Ask whether each one earns its cost at the doubled price. A tool that's marginal at $40 is a cancellation at $80. The AI ROI Reality Check prompt walks a keep-cancel-renegotiate pass over the list.
3. If you're picking a model for a new automation, compare on the January price, not the sticker. Our four-way comparison of ChatGPT, Claude, Gemini, and Copilot covers the capability side of that choice.

One honest caveat: if your entire AI footprint is a chat subscription and no automations, this whole story is a shrug. Bookmark the date and move on.

The pattern behind the price

Google has now shipped two Flash models in three weeks and priced the newer one to pull builders in before ratcheting up. Every frontier lab is doing a version of this: capability keeps climbing while the effective price wobbles around promotions, tiers, and fine print. The takeaway for a small business isn't to chase each release. Pick the assistant you'll show up to daily, build habits on it, and treat per-token costs as a real line item with real renewal dates.

If you want the habit side of that, the SME AI Starter Stack lays out the adoption order that keeps spend boring and output compounding.