Mini, nano, full: three releases this week and the question of which size
OpenAI shipped GPT-4.1 in three sizes on Monday and two reasoning models today. Picking the size is now a real decision with real money in it.
On 14 April 2025 OpenAI released GPT-4.1 as a family — full, mini and nano. Today, 16 April, it released o3 and o4-mini, its newest reasoning models. Five model variants in three days from one provider.
This is no longer a release cadence story. It is a purchasing structure, and every major provider now has the same shape: a large expensive model, a small cheap one, and something in between.
What the sizes actually mean for you
Not intelligence in the abstract. Cost and speed per task, traded against how often the answer is wrong.
The smallest tier is for high-volume, low-judgement work where an error is cheap and visible. Classifying incoming mail. Extracting four fields from a standard form. Tagging documents. These run thousands of times a month and the small model is often indistinguishable from the large one on them.
The middle tier is your default for everything a person reads before using: drafting, summarising, rewriting, translating.
The reasoning tier is for the handful of tasks where you want the answer to be right more than you want it quickly. Contract comparison, planning with real constraints, checking a specification for contradictions. A few times a month, not a few times an hour.
The mistake that costs the most
Using one tier for everything, in either direction.
Running everything on the cheapest model produces a steady trickle of small errors in places nobody is checking. Running everything on the most capable produces a bill that makes the whole programme look unjustifiable at the next budget review.
What to do with this
Take your list of AI-touched tasks — you wrote it in November for the inventory — and put one of three letters next to each: S, M or R. It takes twenty minutes.
Then ask your vendor whether their product lets you set the tier per task or decides for you. If it decides for you, ask how you would find out what it chose. That is the whole conversation, and it is worth having before the invoice rather than after.