Thinking Budget
Thinking Budget
A setting that controls how much computation or reasoning depth an AI uses when generating an answer.
In Simple Terms
A thinking budget is a setting that controls how deeply a generative AI reasons before answering a question. Set a higher budget, and the AI takes its time working carefully through complex puzzles or programming problems. Set a lower budget, and it spends less time thinking so it can respond faster. It's mainly used to balance an AI's response speed against the accuracy of its answers.
Behind the Name
The term treats an AI's reasoning process the same way you'd treat a budget for money: it's a rough guideline for how much you plan to spend, not a hard rule. You might end up spending less than planned, or more. A thinking budget works the same way—it's a guideline for how deeply the AI reasons before producing an answer, not a fixed amount it has to use up.
Take a Closer Look!
A thinking budget is a setting that specifies, as a number, a rough guideline for how many thinking tokens (units of computation) an AI uses in its internal reasoning when working through a difficult problem.
Some AI models with advanced reasoning capabilities work through steps internally before producing their final answer.
A good way to picture it: it's like deciding ahead of time roughly how much scratch work you'll allow yourself for one problem on a test.
The number you set represents a rough guideline for how many thinking tokens the AI uses internally. The AI still adjusts the actual number of tokens it uses based on how hard the problem is—for an easy question, it might finish well before reaching the limit, while for a complex one, it might reason past the number you set. The higher you set the budget, the more room the AI has to think things through, which in turn changes both the response time and the quality of the answer.
A thinking budget is used to balance an AI's response speed against the quality of its answers.
Depending on which AI model supports it and what you're using it for, it's one of the settings that lets you control cost and computing resources while adjusting how the AI produces its answers.