Terms  /  Reasoning Cost  /  Reasoning Load
02 · Reasoning Cost

Reasoning Load

The amount of interpretive work a model must perform before useful task execution begins.

Reasoning Load measures the gap between receiving the prompt and beginning useful work. In an ideal scenario, Reasoning Load is near zero — the model reads the prompt and immediately starts executing the task. In practice, the model often needs to spend significant effort interpreting what it's been asked to do: parsing unstructured instructions, inferring implicit constraints, determining which parts of the context are relevant, and building a mental model of the task. All of this is Reasoning Load — work that precedes the actual work.

Example
A model receives a prompt that includes a 3,000-word project description, six code files, and the instruction "continue where we left off." The Reasoning Load is enormous: the model must read the project description to understand the domain, scan all six files to understand the codebase, and then figure out what "where we left off" means by inferring the last meaningful change. Only after all that interpretation can it begin writing code. A prompt that says "add input validation to the `createUser` function in `auth.ts` — reject emails without @ and passwords under 8 characters" has near-zero Reasoning Load.

More in Reasoning Cost