Open Google's Gemini pricing page and look at an input price:
$0.75 through December 31, 2026.
$1.50 starting January 1, 2027.
Both numbers are in the same cell. Most tools that read that page — including, until recently,
mine — take the first one.
I counted every cell on the page with that shape. There are 70 of them, across 6 models, and
all 70 are exactly ×2.00.
model cells tiers input
Gemini 3.8 Flash 20 Standard/Batch/Flex/Priority/Contrib $0.75 → $1.50
Gemini 3.6 Flash 16 Standard/Batch/Flex/Priority $0.75 → $1.50
Gemini 3.8 Flash TTS 12 Standard/Batch/Flex/Priority $0.50 → $1.00
Gemini 3.8 Flash-Lite TTS 12 Standard/Batch/Flex/Priority $0.50 → $1.00
Gemini Robotics ER 2 Preview 8 Standard/Batch $1.00 → $2.00
Gemini Robotics ER 2 Streaming 2 Standard $1.00 → $2.00
Six of the 32 priced models on the page. Not preview oddities either — Gemini 3.8 Flash is the
one the page calls their most intelligent Flash model, and the doubling applies across all five
service tiers at once, Batch and Flex included.
If you benchmarked a workload against Gemini 3.8 Flash this quarter, the unit economics you wrote
down have about ten weeks left.
Two other vendors did the same thing and it ended differently both times
This construct — a published price with an expiry date on it — is not Google's alone. What
happens at the expiry is where they diverge.
OpenAI states a date and not a price. From their pricing page:
"GPT-5.6 Sol's promotional pricing is available at least through November 21, 2026."
You get a floor, not a schedule. What the price becomes on November 22 is not on the page, and
"at least" means the date is not firm either. That is five weeks out as I write this.
Anthropic stated both, then cancelled the increase. Their pricing page now carries this
footnote:
"The $2/$10 per million input/output token pricing for Claude Sonnet 5, announced at launch as
introductory pricing through August 31, 2026, is now the standard price. The previously
scheduled increase to $3/$15 per million input/output tokens on September 1, 2026 will not
occur."
Same construct, opposite outcome. The introductory price became the permanent one.
So across three vendors you get three results from the same sentence shape: one doubles on
schedule, one won't say, one withdrew the increase. There is nothing in the construct itself
that tells you which you are looking at.
Why your comparison table can't show you this
Here is the part I am least comfortable about, because it is my own site.
My price index renders this:
gpt-5.6-sol — openai · in $4 / out $20
That is correct. It is also the promotional price, and the cell has nowhere to say so. An index
is one number per model per direction. There is no room in that shape for "until November 21,
at least", and the same is true of every price comparison table I have ever seen, including the
ones that scrape mine.
My event archive does carry it — one of the records is literally titled "Google ships two 3.8
TTS models at $0.50 input, doubling on 2027-01-01", because an event is a sentence and a
sentence can hold a date. The table built from those same events cannot. The information exists
on the site and is absent from the page most people would actually look at.
That is a structural limit, not a bug I can go fix this week. The honest version is to say which
surface you are reading.
What I would actually do about it
Check the expiry on the vendor's page, not in a comparison. For any model you have committed
to, open the vendor's own pricing page and read the whole cell. If there are two numbers in it,
your cost model has a date attached to it.
Three dates worth putting in a calendar right now. November 21, 2026 (GPT-5.6 Sol's stated
floor), December 31, 2026 (all 70 Gemini cells above), and — only as a reminder that these can
go the other way — September 1, 2026, the Anthropic increase that never happened.
Don't price a migration off a promotional rate. The Gemini Flash models at $0.75 look
decisively cheaper than several alternatives. At $1.50 the ordering changes, and the switch is
already published. You can make that decision now with the real number instead of being told in
January.
Notes
Read from the vendors' own pricing pages on 2026-10-17: ai.google.dev/gemini-api/docs/pricing,
developers.openai.com/api/docs/pricing, platform.claude.com/docs/en/about-claude/pricing.
The ×2.00 figure is every one of the 70 cells, not a median — input, output, cached input, and
every tier the model is sold under. I checked for exceptions specifically and there are none,
which is itself worth knowing: this is a flat doubling, not a repricing.
Two other vendors I track (xAI, DeepSeek) publish no time-limited language on their pricing pages
at all. That is not the same as a guarantee — it means there is nothing to read.
The archive of what these pages said before, with dates:
aichangewatch.com/price-index
Top comments (0)