Token Budget in plain English.

A token budget is the practical cap a team sets on how many tokens — and therefore how much money — a given AI feature, workflow, or user is allowed to consume per request, per day, or per month. It's a cost-control decision layered on top of a model's technical context window, not a replacement for it: a model might technically support 200,000 tokens of context, but a product team can still decide that any single request should burn no more than 4,000 tokens, or that a free-tier user gets 50,000 tokens a month before a feature throttles or prompts an upgrade. Token budgets matter because API pricing is metered per token, so an AI feature with no budget can quietly turn into an unbounded cost as usage scales. Common tactics for staying inside a budget include summarizing long context instead of sending it in full, routing simple requests to a cheaper model, and capping how much conversation history carries forward — all standard tools of the wider discipline of context engineering.