Enabling Auto mode quietly turns on a second LLM (the safety classifier) that increases your token usage by ~15-28%, and Anthropic used to charge you for these

Enabling Auto mode quietly turns on a second LLM (the safety classifier) that increases your token usage by ~15-28%, and Anthropic used to charge you for these tokens. Per this announcement, they've stopped charging for the tokens used by this safety classifier.