Why reserve completion tokens up front?

The context window is shared between input and output. If you do not carve out room for the reply, the model has no space to answer and the call truncates or errors. Reserve generously for reasoning models.

How much should I leave for retrieved context?

Whatever is left after system prompt, user message, and reserved completion. This tool computes that remainder so you know exactly how many tokens of RAG chunks you can fit.

What if the fixed slots already exceed the window?

The tool flags an overflow and shows the negative remainder, meaning you must shrink the system prompt, trim the user message, reduce the reserved completion, or move to a larger-context model.

It is exact arithmetic on the token figures you provide. Estimate those figures with a token counter, and leave ~10-20% margin for estimate error and special tokens.

Token Budget Splitter

Token budget splitter

A model’s context window is a fixed budget you have to divide between four competing demands: system instructions, retrieved context, the user’s message, and room for the reply. This tool takes your context limit and fixed slots, then tells you exactly how many tokens are left for retrieved context chunks — the slot that usually flexes in RAG systems.

How it works

You enter the total context window and three known quantities: system prompt tokens, reserved completion tokens, and your typical user message size. The splitter subtracts those from the window and reports the remaining budget available for retrieved context, plus a percentage breakdown of how the whole window is allocated. If your fixed slots already exceed the window, it flags the overflow and shows by how much you are over.

Tips and notes

This is the core sizing step for any RAG pipeline: it tells you how many chunks (or how large a chunk) you can retrieve before you blow the budget. Reserve completion tokens generously — reasoning models emit hidden thinking tokens that count against the window. Leave 10-20% slack for token-estimate error and special tokens rather than packing to 100%. To validate a specific assembled prompt against a model, pair this with the context window planner.