Google's Gemini 4 Carbon API doubles in price December 1. Run your numbers this week.
Introductory pricing of $2 per million input tokens and $10 per million output tokens is scheduled to go to $4 and $20 on December 1. Here's how to calculate what that does to your bill before it lands.
By PANONDA Newsroom

Foto: Markus Winkler · Pexels
What happened
Google launched Gemini 4 Carbon with introductory API pricing: $2 per million input tokens and $10 per million output tokens, good through November 30, 2026.
On December 1, those rates are scheduled to double — to $4 per million input tokens and $20 per million output tokens.
That is the whole story. No new capability, no deprecation, no migration deadline. Just the end of a launch discount that a lot of people built workflows on top of without noting the expiry date.
Who this actually hits
If you use Gemini through a chat subscription, nothing changes for you. These are API rates — the per-token price you pay when software calls the model.
The people affected are publishers running the model inside something: archive search, transcript cleanup, show-notes generation, headline variants, comment moderation, metadata and tagging, newsletter drafting, or any reader-facing feature wired to a Gemini endpoint. If you pay a monthly bill to Google Cloud or the Gemini API and the line item is tokens, read on.
The math is unforgiving because it's linear: if your token volume stays flat, your bill exactly doubles.
The formula is simple enough to do in a spreadsheet. Input tokens divided by a million, times the input rate, plus output tokens divided by a million, times the output rate.
A workload burning 50 million input tokens and 10 million output tokens a month pays 50 × $2 + 10 × $10 = $200 today. On the new rates: 50 × $4 + 10 × $20 = $400. That's $200 more a month, $2,400 a year, for the identical work.
Output is where the pain concentrates. At $20 per million, a model that writes long answers costs five times what a model that classifies costs. If you're generating full drafts rather than short labels, your output column is the one to attack first.
What this story is not
It is not confirmed that every Google surface moves on the same date at the same rate. The available reporting does not establish whether the Gemini API, Vertex AI and enterprise contracts all land on $4/$20 on December 1, or whether existing customers get any grandfathering. If you're on a negotiated contract, ask your rep in writing before you replan anything.
It is also not a signal that Google is turning expensive across the board. Introductory pricing ending is the normal shape of a frontier-model launch. Batch and cached-input discounts are advertised elsewhere in Google's pricing — industry comparisons cite a 50% batch discount — but whether those apply to Carbon on the new rates should be confirmed on your own console, not from a comparison blog.
And the obvious switch isn't automatically cheaper. OpenAI reportedly cut GPT-6.2 Astra by 20%, to $8 per million input and $40 per million output, according to secondary reports. That's still above Gemini's post-increase rate on both sides. Cheaper frontier model does not mean cheaper than this frontier model.
Do this before Monday
- Export your last 30 days of Gemini token usage, broken out by model and by project. You cannot plan without the input/output split.
- Multiply that usage by the December rates. Write the number down. That is your new baseline, not a worst case.
- Separate experiments from production. Kill any test harness still looping against the API that nobody remembers starting.
- Move offline, non-urgent jobs — archive backfills, bulk tagging, nightly summaries — to batch processing if your workload qualifies. Confirm the discount applies to Carbon on your own billing page.
- Downgrade the routine work. Tagging, short summaries, metadata and classification rarely need a frontier model. Run a week of A/B output on a cheaper tier before you decide.
- Cache repeated system prompts and reference documents wherever the API supports it. Cached input is the cheapest token you'll buy.
- Set a hard spending alert dated before December 1, so the first doubled invoice isn't how you find out.
- If you're on an enterprise or Vertex contract, email your account rep and get the December rate in writing this week.
Sources
Everything above was checked against these pages. Open them and see for yourself — that is why they are here.
Get the stories by email
What changes in AI for people who publish — sourced, and to the point.
No spam. Unsubscribe anytime.