loading
loading
The cap on how many tokens a task gets to spend before it must finish or get cut off.
Every request to a model costs tokens for both input and output, and a token budget is the ceiling you set on that spend for a single task, request, or session. It matters because long agent loops can silently balloon costs and blow past the context window, so you allocate a budget and trim history, summarize, or stop when you hit it. Example: an agent gets 50k tokens per run; once it's used 40k on tool calls and reasoning, you force it to produce a final answer instead of looping again.
Plainly
Think of Token Budget as the checklist that keeps the app open for real people. The cap on how many tokens a task gets to spend before it must finish or get cut off.
In practice
Use it when a change has to survive deploys, users, incidents, analytics, or billing reality. In practice, define the owner, input, output, and failure mode before you rely on it.