GPT-4.1 made coding models feel specialized

GPT-4.1 made coding models feel like tools for different jobs rather than one generic score.

Full, mini, and nano made routing visible. A repository repair, a migration review, a diff summary, and a typo fix have different failure costs.

job	route	why
explain a test failure	mini	cheap read of logs and stack traces
patch a tricky bug	full	broader context and reasoning
summarize a diff	nano	narrow and low risk
review a migration	full	expensive if wrong

The route should match the job before it spends the larger budget.

Long context helps only when patch discipline survives it. A model that can see more of the repository still needs to keep the change small, respect surrounding files, and leave a readable working tree. More context should not turn a code editor into a rewrite bot.

The practical difference is straightforward. A typo fix can return a one-file patch through the cheap route. A schema migration may need the full model because its neighboring changes matter. A diff summary can use the smallest route and produce a review note.

Specialization is useful when the product can name the work, choose the route, and verify the result instead of sending every coding request to the largest model.