I'm very confused by Anthropic's claims... They claim Sonnet 5.5 uses much less tokens and therefore has a lower cost per task than Sonnet 5. However, Artificial Analysis has it basically being the most token-inefficient model out right now. What am I missing?
