The default model just got smaller, and that is the year’s real story
Google made its fast, cheap model the default in its assistant yesterday. After a year of ever-larger launches, the direction has quietly reversed.
On 17 December 2025 Google released Gemini 3 Flash, the fast and low-cost member of the family, and made it the default model in its assistant. The largest model is still there; it is no longer what most people get.
That is the most interesting decision of the month, and it closes a theme this blog has followed since February.
Why a provider makes the small model the default
Because most questions do not need the large one, and serving them with it is expensive for the provider and slower for the user. The 16 October argument — that most business work does not need the largest model — is now the industry’s own operating assumption.
It also follows the 25 September piece on ten gigawatts: when capacity is the binding constraint, routing cheaply is not a cost optimisation, it is how the service stays available at all.
What this means for your own choices
Two things. First, if you pay per use and have not looked at which tier your traffic lands in, the industry has just told you where the money is: the 8 August question about a per-tier report is more answerable now, because providers are building exactly that plumbing for themselves.
Second, the small-model test from 16 October is worth actually running in January. Take your two most frequent tasks, run them on the cheapest tier for a week, and see whether anyone notices. In our experience one of the two moves down permanently.
The year in one sentence
January to August was about models getting bigger; September to December has been about getting the right-sized model to the right request.
The second half is better news for a company of thirty people than the first, and it received perhaps a twentieth of the coverage.