#ReasoningModels

Local Reasoning Model Token Cost — thinking ON/OFF Measured

I ran 13 questions on gemma4:12b with thinking ON and OFF. Reasoning got one more right while spending 68x the output tokens and 19x the wall-clock.