Reasoning tokens
Definition
Tokens used internally by some reasoning models while working through a problem before producing the visible answer.
Put simply
The AI can consume computing effort that you never see in the final response.
What boards and leaders should understand
A concise-looking answer does not necessarily mean little work or low cost occurred behind it.
Complex reasoning can consume additional tokens internally, which is one reason AI economics cannot always be inferred from the length of the visible output.
Questions to consider
- For reasoning-intensive workloads, do we understand what is driving cost rather than simply looking at the visible output?
