Study: Task completion rates matter more than token costs in AI pricing

Databricks found Claude Code used 742,000 context tokens per task versus 236,999 for Pi on Opus. The Register argues token-based pricing is misleading and task completion rate is a better cost metric.