Gemini 3.7 Flash's Intro Price Doubles in 100 Days. Here's What That Actually Costs You

22 September 2026 · AI & LLMs · 5 min read

Google shipped Gemini 3.7 Flash on September 2 at $0.75 per million input tokens and $3.75 per million output — cheap enough that plenty of teams wired it straight into production without reading the fine print under the price table. That fine print says the rate holds "through December 31, 2026." Starting January 1, 2027 — 100 days from today — the exact same model, same weights, same output, bills $1.50/$7.50. A clean doubling, on both input and output, with no model swap and no separate announcement beyond a line in Google's own pricing docs. I checked that rate card directly, matched it against our own tracked pricing on apicostcalc.com/gemini.html, and ran two real workloads through both sides of the deadline to see what the doubling actually adds to a bill.

The rate card, before and after

Only Gemini 3.7 Flash carries this specific expiry — Google's own docs flag it as an introductory rate, not the other 3.x models. It's the same mechanism as a promotional launch rate on a SaaS plan: get adoption in the door cheap, then step up to the number the model was always going to cost once the free-trial period on it ends. Nothing about the model changes on January 1 — same context window, same latency class, same benchmark scores. Only the invoice does.

Workload 1: a moderation or classification batch job

A content-moderation pipeline tagging 200M input tokens and generating 20M output tokens a month — a typical shape for a job that reads a lot of text and writes short labels back.

An extra $225 a month for a job that's easy to underestimate precisely because $225 doesn't sound like a lot. It's still an exact doubling — the same multiplier applies whether the base bill is $225 or $225,000.

Workload 2: a RAG app at real request volume

A retrieval-augmented app serving 500,000 requests a month, each pulling roughly 3,000 tokens of retrieved context plus the user's query, and writing back a 300-token answer. That's 1,500M input tokens and 150M output tokens a month — a workload big enough that this deadline shows up as a real budget line, not a rounding error.

That's $1,687.50 more every month, or just over $20,000 more a year, from a pricing-docs line item rather than a model change, a feature launch, or a usage spike. If this workload is already running on 3.7 Flash, that number belongs in a Q1 2027 budget now, not discovered on the January invoice.

What moving to Flash-Lite instead would actually save

Gemini 3.1 Flash-Lite bills $0.25/$1.50 with no expiry flagged in Google's docs. Run the same RAG workload's token volumes — 1,500M input, 150M output — through Flash-Lite instead:

That's a 5.6x gap, and it's tempting to read this as "just switch." I wouldn't, not blindly. Flash-Lite is a smaller, cheaper-tier model than Flash — it trades away some of the reasoning and instruction-following headroom that made 3.7 Flash worth picking over it in the first place. The honest move is to run your own eval set through Flash-Lite before the deadline and see whether the answer quality holds up for your specific retrieval-and-answer task. If it does, you bank the $2,775 a month. If it doesn't, you now know that with data instead of guessing, and you budget for the doubled Flash rate with your eyes open.

Что бы я на самом деле сделал

Anything you've shipped on Gemini 3.7 Flash needs a line in the January 2027 budget today — the rate change is already public, so there's no ambiguity to wait out. Before that date, run your real workload against Flash-Lite and measure the quality delta instead of assuming it either "will obviously be fine" or "obviously won't work." And treat every introductory AI API rate the same way going forward: the discounted number gets you evaluating the model, not a permanent price — check the provider's docs for an expiry line before you build a cost model around launch-week pricing.

• Калькулятор стоимости Gemini API — plug in your own token volumes and compare the current and post-January rates side by side.
• Gemini API pricing & how to get a key — the full current rate card for every Gemini model, kept in sync with Google's docs.
• Калькулятор стоимости Тряпки — model your own retrieval-plus-generation workload instead of the 500K-request example above.
• Цены на OpenAI, Claude и Gemini — see where Flash-Lite and post-hike Flash land against the other providers' cheap tiers.

Новичок в расценках LLM на основе использования? Начните с бесплатные руководства по стоимости API.

Разверните его самостоятельно: DigitalOcean — бесплатный кредит в размере 200 долларов США ↗ · Хостингер VPS ↗

Pricing verified against Google's official Gemini API pricing docs (ai.google.dev/gemini-api/docs/pricing) and our own live rate tracking on apicostcalc.com/gemini.html, checked 2026-09-22. Gemini 3.7 Flash launched September 2, 2026. Reference estimates using Google's published pay-as-you-go rates — Vertex AI pricing, enterprise agreements and future rate changes vary; confirm current pricing at the source before budgeting. Workload figures (request volumes, token counts) are illustrative scenarios computed against real published per-token rates, not vendor-supplied numbers.