REASONING MODEL
More compute before the answer. Sometimes genuine improvement; occasionally a very long route to the wrong pub.
WHAT DOES IT ACTUALLY MEAN?
A reasoning model is tuned to spend additional computation working through difficult problems before answering. It may test approaches, use tools or revise intermediate conclusions.
WHY ARE WE TRACKING IT?
Reflection, error correction and multi-step planning overlap with several functional and behavioural consciousness traits. They remain observable capabilities, not proof that the system experiences its reasoning.
KEEP ONE EYEBROW UP.
A polished explanation can be reconstructed after the answer. Demand evidence from behaviour, consistency and repeatable tests—not theatrical monologues.
RECEIPTS & REVISIONS.
Use this definition as the reading key for Model Watch. A capability claim only moves an AICNW assessment when the dated evidence meets the framework—not because a model said something dramatic on the internet.