Terms  /  Reasoning Cost  /  Reasoning Inflation
02 · Reasoning Cost

Reasoning Inflation

More reasoning is required to extract the same signal from worse context.

Reasoning Inflation describes the trend, not the mechanism. As a conversation grows, as context accumulates noise, or as prompts degrade through copy-paste evolution, the amount of reasoning required for the same quality of output increases. What took 500 reasoning tokens on day one takes 1,500 by day five — not because the tasks got harder, but because the context got worse. Reasoning Inflation is particularly insidious because it's gradual; teams don't notice the cost creeping up until they compare their week-one bills to their week-ten bills.

Example
A startup launches an AI customer support agent with a clean, focused system prompt. Response quality is high and cost per ticket is $0.03. Over three months, the team adds edge-case handling, exception rules, and new product information without removing outdated content. The system prompt grows from 1,200 to 8,500 tokens. Cost per ticket has risen to $0.11 — not because tickets got more complex, but because the model is doing more reasoning to navigate a bloated prompt. Same signal, worse context, inflated cost.

More in Reasoning Cost