1 nota · última: 9 Jul 2026
The post argues that per-token pricing is a poor way to compare AI models because tokenizers differ and token efficiency varies. It favors cost per completed task as the more useful metric.