Gemini 4 Carbon costs $2 per million input tokens — until November 30
Google opened its more capable Gemini tier to every paying developer at the cheaper tier's price. A pricing tracker says the rate doubles after November 30. Here's what to test while the discount holds.
By PANONDA Newsroom

Foto: Markus Winkler · Pexels
What happened
Google opened Gemini 4 Carbon to any paying developer on November 10. The listed price is $2 per million input tokens and $10 per million output tokens — the same rate as the cheaper Gemini 4 Argon tier, according to a November pricing report.
A separate pricing tracker says that rate is promotional and is scheduled to rise to $4 per million input tokens and $20 per million output tokens after November 30.
Google's own announcement page and the official end date could not be confirmed from the material we have. Treat the November 30 cutoff as reported, not as a Google commitment.
What the numbers actually mean for a publishing budget
Token pricing is abstract until you anchor it to a job.
Say you run 10 million input tokens and 1 million output tokens through the model in a month — roughly the shape of summarizing a mid-size archive, generating metadata, and drafting descriptions at volume. At the current rate that's about $30. At the reported post-November rate it's about $60.
That's before storage, orchestration, retries, and whatever your platform charges on top. The point isn't that $30 versus $60 breaks anyone. The point is that you get three weeks to find out what this class of model does to your workflow at half the price you'd otherwise pay for it.
What this story is not
It is not a claim that Carbon is better than what you're using. The available material confirms the price and the access change. It does not confirm benchmark results, context window size, output limits, rate limits, geographic availability, or whether cached-input and batch discounts apply to Carbon the way they do to other tiers.
It is also not a reason to migrate a production workflow this month. A promotional rate that reportedly doubles in three weeks is a bad foundation for anything you can't unwind. The useful move is a test, not a switch.
And it is not the only price move in the window. Anthropic's Fable 5.2 is listed at $8 per million input and $40 per million output, with batch processing at $4/$20 and cached reads at $0.20 per million, according to the same round of tracker reporting — figures we could not independently verify. If you're pricing a job, price it in at least two places.
The comparison that matters
Small publishers tend to ask which model is smartest. That's the wrong question for this kind of work.
The right one: which model produces output you'd actually ship — or actually use internally — at the lowest verified cost per unit of work? A model that's 10% better and 4x the price loses on transcript cleanup. It may win on a legal summary where an error costs you a correction.
Run that comparison now, with a stopwatch and a spreadsheet, while one side of it is discounted.
What to do this week
- Pick one repeatable job — archive summaries, episode descriptions, structured metadata, transcript cleanup. One job, not five.
- Run 50 real items through Carbon and through whatever you use now. Same prompt, same inputs.
- Score factual errors by hand. Count them. A cost comparison without an error rate is not a comparison.
- Log your exact token usage for the test. You'll need the baseline to model the reported price increase.
- Calculate cost per usable output, not cost per million tokens. Three rewrites at $2 is more expensive than one at $8.
- Check Google's official API pricing page yourself before you plan around November 30. The cutoff is reported by a tracker, not confirmed by Google in what we've seen.
- Keep the test off your publishing path. Internal tooling first. Decide on publication-facing use after you have the error numbers.
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.