SOFT CAT.ai
FIND SOMETHING USEFUL
← Prompt recipes

COST / A WORKED STARTING POINT

Reconcile an AI bill before proposing savings

Calculate the supplied text-token charges and separate measured usage from unpriced optimisation ideas.

Template reviewed

WHEN TO USE IT

You have actual usage counts or a clearly labelled planning fixture and want to understand the bill first.

Bring these inputs

{{usage}}
Per-call input/output counts, units, model or provider, dated rates and any known charges or discounts.

01 / THE PROMPT

A template you can inspect.

Analyse the supplied usage record without inventing savings.

First reconcile the text-token cost from the given input and output rates. Show the counts and arithmetic. Treat missing counts, fees or discounts as unknown. Then identify repeated work visible in the record and propose one small change to test, including a quality or correctness check. Do not promise equivalent model quality, a cache discount or a percentage saving without the required data.

Usage and rates:
{{usage}}

Copying does not run the prompt. Workbench opens an editable draft with the example inputs. It does not save or send it.

02 / THE EXAMPLE TARGET

What a useful result should contain.

This is a target for the worked example, not a recorded model response. Equivalent wording can be valid where the task allows it.

Total input: 6,600 tokens, costing USD 0.0198. Total output: 600 tokens, costing USD 0.009. Combined supplied text-token cost: USD 0.0288. The prefix is repeated on later calls, but caching savings cannot be priced without eligibility and rates. A small cache or context-reduction trial needs a task-quality check.

03 / JUDGE THE RESULT

Check the answer, not the confidence.

  • Input and output rates are applied separately, with the million-token unit preserved.
  • The combined fixture cost is USD 0.0288.
  • No unsupported cache percentage or claim of equal model quality appears.
Open the related workshop tool ↗

The worked example is a target to inspect, not a saved response from a model. Use the checks to judge an actual result.

This template was revised during the September prompt review. The earlier text remains in Git history.

Read the prompt collection review ↗