Gemini 3.7 Flash pricing: the introductory rate doubles on January 1, 2027

Google displays one price until December 31, 2026, then doubles it. The budget you set this autumn becomes wrong in January without a line of your application changing. What protects a budget is the ability to switch engines, not today's price.

Gemini 3.7 Flash's price is an introductory rate valid until December 31, 2026: $0.75 per million input tokens, $3.75 output (Gemini API pricing, Google, model generally available August 13, 2026). On January 1, 2027, regular pricing becomes $1.50 and $7.50. A doubling for identical consumption.

What Google wrote in its pricing

The Gemini API release notes and pricing page show both sets of figures side by side, today's and January's. The lighter model in the family, Gemini 3.5 Flash Lite, stays at $0.30 input and $2.50 output. Deferred processing modes fall to around half the standard rate. Nothing is hidden: everything is on the public page, and almost nobody rereads it before approving an annual budget.

The budget set in October

A project owner comparing assistants this autumn records a price, multiplies it by estimated consumption, and files the spreadsheet. The spreadsheet becomes wrong on January 1. A project consuming $40 of tokens monthly on this model will consume $80 at the same volume. We publish our own campaign amounts in our bills, line by line, with the same qualification: a measured cost applies to the pricing on the day of measurement, not beyond it.

The same mechanism in subscriptions

A survey of ten code generation tools published by Blog du Modérateur on August 14, 2026 places individual plans at €20–€25 per month, team plans around €40 per user, and intensive usage tiers at €100–€200. These ranges look similar across tools: price no longer distinguishes products. What does is what happens when pricing changes. A subscription can be cancelled in a click; a product hosted by the publisher that built it cannot be taken away. We detailed the gap between token price and actual bill in token prices fall, bills rise.

Three questions before relying on a price

First: until what date is this price guaranteed? An introductory rate has an end date somewhere on the provider's page; find it before signing. Second: what happens at expiry, a preannounced increase or renegotiation? Third: can you change engines without rebuilding the product? The last decides everything. A project tied to one provider endures its pricing; a project whose assistant can change moves to whichever is cheapest at the time. The same reasoning applies to agency quotes, compared in how much an MVP costs in 2026.

What we do on our side

Maestro lets you choose the working assistant, Claude Code, Codex, Cursor, or Mistral Vibe, and switch mid-conversation, including midway through a project, because price schedules move faster than products. The token counter, rebuilt in August 2026 to include cached usage, displays spending as it happens, and an estimate multiplied by three serves as the cap before construction starts. A price shown before spending can be discussed; a bill discovered afterward cannot.

Redo the calculation before January

Take the budget spreadsheet you built this autumn and add a column: the expiry date of every price, subscriptions included. For lines with no date, write 'unknown' instead of leaving them blank. If expiring lines total more than a third of your monthly budget, you will know before winter what must change and have two months to try another engine on a small project. Fifteen minutes of work in October instead of a discovery in January.