Google Has Pre-Announced a 2x Gemini Flash Price Rise. Budget for It Now

Google's own pricing page, captured on our desk on 1 September 2026, states that Gemini 3.7 Flash model prices double on 1 January 2027: input from $0.75 to $1.50 per 1M tokens, and output from $3.75 to $7.50 per 1M tokens. Context caching and storage prices double on the same date.

That is the verified layer, and it is unusually clean. It comes from the vendor's own pricing page, it carries a date, and it names the exact figures. Most price changes in this category arrive as a quiet page edit that someone notices six weeks later on an invoice.

The same capture also records that the Gemini API now runs three tiers — Free, Paid and Enterprise — with the Enterprise tier adding dedicated support, security and compliance features, and volume discounts. If you are large enough to be affected by the Flash rise, you are probably large enough to ask what the volume discount looks like.

Most of the cost damage we see on this site comes from changes nobody planned for: a plan repackaged mid-quarter, a credit system replacing a seat, a model deprecated into a dearer successor. Those are hard because the decision and the discovery happen at the same moment.

This one is the opposite. The rate change is dated, the figures are published, and the date is far enough out that you can act deliberately. There is no emergency here. There is only the risk of doing nothing until there is.

Here is the uncomfortable part. Teams that cannot answer "what share of our tokens is on Flash?" in under a day will also not act on this in time. The announcement is a good excuse to fix that instrumentation gap, which is worth more than the price change itself.

The exercise is small and you can do it this week. Take last month's Flash token counts, split input and output, and price them twice: once at today's $0.75 / $3.75, once at the announced $1.50 / $7.50.

The gap between those two numbers is your exposure. If it is under a few hundred dollars a month, note it and move on. If it is a meaningful share of your AI line, you now have a named, dated problem with four months of runway — which is a far better position than most cost problems arrive in.

Do the same for context caching, which doubles too. Cached-input strategies that were justified by a cheap cache tier need re-checking at the new rate; the caching maths can flip when the cache itself gets dearer.

Our catalogue currently carries high-throughput rows well below Flash. That is not a recommendation to move blindly — a cheaper row that fails review twice is dearer than an expensive row that passes once. It is a reminder that Flash is not the floor and never was.

The test is boring and it works. Take a representative sample of the traffic you would move — a few hundred real requests, not synthetic ones. Run them on the candidate row. Have a human score only whether the output would have been accepted. If acceptance holds, the move is real. If it drops, you have priced the difference and you keep Flash with a clear conscience.

At the end of the day, the model that costs less per accepted outcome wins, and that is rarely the same as the model with the lowest number on the price sheet.

Do not rip out Flash in September for a price that changes in January. You would pay migration cost now to avoid a cost later, which only makes sense if the migration is cheap and the exposure is large.

Do not assume the rise will be reversed or quietly delayed. Plan against the published date. If Google moves it, you have lost nothing but a spreadsheet afternoon.

Do not let this become a general "Google is getting expensive" narrative in your organisation. One model tier's rate is doubling on a named date. That is a specific, boundable fact, and treating it as a vendor-wide verdict is how teams end up paying migration costs three times.

Best move for most teams: keep Flash, re-baseline the 2027 budget at $1.50 / $7.50 now, and instrument your token split by model so this is a five-minute question next time.

Best move for high-volume bulk pipelines: run the acceptance test on a cheaper catalogue row this quarter. If it passes, move before January and take the saving twice — once at today's rate and again at the new one.

Best move if you are on the Paid tier at real volume: ask Google what the Enterprise tier's volume discount does to the 2027 number before you plan anything else. A published list price is the start of that conversation, not the end.

Avoid: a panic migration in September, and a budget built on the old rate. Confirm the current figures on Google's pricing page before you commit — this page reports what that page said on 1 September 2026.