Why doesn't doubling context double my cost?

Because only the input (context) portion of a request scales with context size. Output tokens are billed separately and stay fixed, so if output is a meaningful share of each call, doubling the context grows the cost by less than 2x.

When would doubling context nearly double cost?

When your output is tiny relative to your input — for example a long RAG context with a one-word classification answer. There the bill is almost all input, so the cost ratio approaches the context ratio.

Does a bigger context window itself cost more?

A model's maximum context capacity (e.g. 128K vs 1M) is a capability, not a charge — you only pay for the tokens you actually send. Some providers do price very long contexts at a higher per-token tier, which you can reflect by editing the input price.

Should I always use less context to save money?

Not necessarily. More context can improve accuracy and cut expensive retries. Use this tool alongside the retrieval-quality tradeoff tool to balance cost against output quality.

Is my data sent anywhere?

No. The calculation runs entirely in your browser. Nothing you enter is uploaded, stored or logged.

Context Window Doubling Cost Calculator

Doubling the context does not double the bill

A common budgeting mistake is assuming that sending twice as much context costs twice as much. It almost never does. LLMs bill input and output tokens separately, and only the input portion grows when you enlarge the context — your output stays the same length. This calculator shows the real cost change between your current and proposed context sizes for any model and volume.

How it works

Each request costs (context_tokens × input_price) + (output_tokens × output_price). When you raise the context from, say, 4,000 to 8,000 tokens, only the first term doubles; the output term is unchanged. So if output is a meaningful slice of each call, your cost ratio comes in below the context ratio. The tool reports both ratios explicitly, along with the per-call and monthly cost before and after, so you can see exactly where the money goes.

Tips for sizing context economically

Watch your output share. The bigger your output relative to input, the cheaper extra context is in percentage terms.
Cache the stable part. If most of your context is a fixed system prompt or document, prompt caching can cut the input cost of the repeated portion sharply.
Trim before you grow. Remove redundant or low-value context first; it is cheaper than paying for a larger window full of filler.
Balance cost and quality. More context is not always better — see the context-size-vs-retrieval-quality tool for the diminishing-returns curve.