Cheap Reasoning Arrived in January and Reset the Cost Model

January’s arrival of a competitive open-weight reasoning model at a small fraction of prevailing API prices did two things at once: it moved markets, and it moved the floor under everyone’s pricing.

What it meant for people building products

Reasoning capability had been the expensive tier reserved for tasks that justified it. Suddenly a multi-step reasoning call was cheap enough to put in an ordinary workflow, and the calculus on a lot of shelved features changed overnight.

Open weights change the deployment options too

Self-hosting becomes viable for organizations that cannot send data to a third-party API at all. That is not most teams — running inference well is real operational work — but for regulated environments it converts a hard no into a costed decision.

Any capability that is expensive today is a capability worth designing for as though it will be cheap next year.