Terms  /  Reasoning Cost  /  Reasoning Tax
02 · Reasoning Cost

Reasoning Tax

Extra model cost paid to compensate for context debt.

Reasoning Tax is the most direct connection between context quality and your invoice. When a model receives a messy prompt, it doesn't simply fail — it tries harder. It spends additional reasoning tokens interpreting ambiguity, resolving contradictions, and filtering noise before it can begin the actual task. This extra reasoning appears on your bill as higher token consumption, but it produces no additional value. You're paying more for the same result you'd get from a clean prompt — or often, a worse result at a higher price.

Example
A team sends a 6,000-token prompt with conflicting instructions to generate a database schema. The model uses 3,200 reasoning tokens to produce the output — 1,800 of which were spent reconciling the contradictions and interpreting vague requirements. After the team cleans up the prompt to 2,400 focused tokens, the same task completes with 1,100 reasoning tokens and higher accuracy. The 1,800-token difference was pure Reasoning Tax — cost with no value.

More in Reasoning Cost