AI cost depends on task success, not token price alone

In a measurement, Claude Code used 742,000 for each task. Pi used 236,999 while running the same model. By token use alone, Pi consumed roughly one-third as much.

A useful cost comparison must also include how often each tool completes the task successfully. A tool that uses fewer tokens can still cost more overall if failures lead to repeated attempts.

Key points

  • Claude Code used 742,000 per task.
  • Pi used 236,999 tokens per task with .
  • Pi consumed roughly one-third as many tokens as Claude Code.
  • Cost comparisons should include rates and retry costs.
Read original