Measure the Cost of an Accepted Result

The cheapest Claude Code session can be the expensive one if its output takes another hour to repair. Compare accepted results, not just token totals.

Choose a repeatable task, such as adding a validation rule to a small import script. Record the session's reported usage, elapsed time, checks passed and your correction time. Use the same starting revision and acceptance conditions for the next attempt.

Do not turn two runs into a universal claim about a model or workflow. Different code, context and retries can change the result. Keep a small table of representative tasks instead. The decision you want to support is practical: which setup produces work you can accept at a cost and delay that fit this job?

Claude Code documentation.