What Is Reasoning?
Reasoning refers to a chain-of-thought process: the model formulates explicit intermediate steps before arriving at the final answer. This makes the solution path traceable and improves quality on demanding tasks.
Reasoning Tokens
Section titled “Reasoning Tokens”During the reasoning process, the model generates so-called reasoning tokens. These are delivered to customers and billed at the same rate as output tokens.
Reasoning Levels
Section titled “Reasoning Levels”Not every model has reasoning capability. For models like GPT-OSS, reasoning effort can be set in levels: low, medium, and high.
Benefits
Section titled “Benefits”- Better quality on complex tasks (math, logic, code generation)
- Traceable solution path
Drawbacks
Section titled “Drawbacks”Recommendation
Section titled “Recommendation”For models with adjustable reasoning effort, choose the level that fits the task: high reasoning levels pay off where complex tasks justify the extra tokens, for simple tasks, lower levels save cost and latency.